A recent advance in the science of chemical separations known as 'comprehensive two-dimensional gas chromatography,' or GC X GC, routinely separates 2000 chemical species from petroleum derived mixtures such as gasoline and diesel fuels. The separated substances are observed to fall into orderly patterns in a two-dimensional image representative of compound classes and isomeric structures. To interpret these complex images, two procedures are needed. First, the images must be transformed into a standard format that permits facile recognition of chromatographic features. Second, quantitative data must be extracted from designated features. By automating these procedures, it becomes possible to rapidly interpret very complex chemical separations both qualitatively and quantitatively.

Determining position, attitude, and aimpoint of a small smart weapon as a function of time is extremely difficult and expensive. This paper describes an accurate, low-cost, portable method for collecting flight imagery and extracting position, attitude, and aimpoint information. This system consists of an onboard CCD camera, transmitter, and associated ground-based equipment. A firepulse from the weapon computer asynchronously takes a snapshot of the aimpoint. The ground- based equipment consists of video cassette recorders and a real-time digital disk recorder as well as GPS-derived IRIG-B encoders. Once the 60 Hz imagery is received and encoded with IRIG-B timecode, it is recorded. The stored imagery is then transferred to a computer workstation for digital image processing that includes individual field and line correction to remove electronic distortions and timing ambiguities. Optical distortions are corrected pixel-by-pixel using pixel maps to 'rubber-sheet' the imagery or build camera models. Images, grouped into several blocks, may be imported directly into Orthomax^R for further processing, which includes manually tagging fiducials and ground control points so that links between image and real space can be made and a 'least squares bundle adjustment' can be applied. An alternative method uses the City University Bundle Adjustment program to perform the necessary mathematical computations on all images in one large data set. An automated method, Video Motion Modeling System, is used to significantly reduce the labor and time involved in data processing. Results from all three methods are presented and compared in the following sections of this paper.

This paper describes a Moire fringe based image analysis system that was developed to determine the out-of-plane deformation of target plates impacted by high velocity projectiles. Due to the highly dynamic conditions that occur as the results of these impacts, such data are very difficult to acquire and analyze. Nevertheless, they are essential for the evaluation of armor materials designed to minimize behind- armor debris, and to study the fracture effects of very strong yet brittle advanced armor materials. Additionally, the data sought is required for fundamental verification of computational material models meant to simulate failure in ballistic experiments. The major goal for developing such a system was therefore to image and analyze the three- dimensional high-speed deformation, fracturing, and propagation of fractures that leads to the onset of fragmentation of targets during impact. The specific image processing methodologies discussed include noise reduction, automated and assisted Moire fringe finding, and the remapping of two-dimensional fringe patterns into three-dimensionally distorted surfaces. Results of applying the image processing system are also provided and methods for increasing system robustness in the presence of higher levels of noise are also discussed.

This paper presents scale invariant face detection and classification methods which use spectral features extracted from Log-Polar image. Scale changes of a face in an image are represented as shift along the vertical axis in Log-Polar image. In order to make them robust to the scale changes of faces, spectral features are extracted from the each row of the Log-Polar image. Autocorrelations, Fourier power spectrum, and PARCOR coefficients are used as spectral features. Then these features are combined with simple classification methods based on the Linear Discriminant Analysis to realize scale invariant face detection and classification. The effectiveness of the proposed face detection method is confirmed by the experiment using the face images which are captured under the different scales, backgrounds, illuminations, and dates. We have also performed the experiments to evaluate the proposed face classification method using 2800 face images with 7 scales under 2 different backgrounds.

PAR Government Systems Corporation (PGSC) has recently developed a complete activity detection and tracking system using standard NTSC Video or Infrared (IR) camera inputs. The inputs are processed using state-of-the-art signal processing hardware and software developed specifically for real-time applications. The system automatically detects and tracks moving objects in video or infrared imagery. Algorithms to automatically detect and track moving objects were implemented and ported to a C80 based DSP board for real-time operation. The real-time embedded software performs: (1) Video/IR frame registration to compensate for sensor motion, jitter, and panning; (2) Moving target detection and track formation; (3) Symbology overlays. The hardware components are PC based COGS which include a high speed DSP board for real-time video/IR data collection and processing. The system can be used for a variety of detection and tracking purposes including border surveillance, perimeter surveillance including building, airport, correctional facilities, and other areas requiring detection and tracking of intruders. The system was designed, built and tested in 1998 by PAR Government Systems Corporation, La Jolla, CA. This paper addresses the algorithms (Registration, Tracking, Outputs) as well as hardware used to port the algorithms (C80 DSP board) for real-time processing.

In this work two methods are presented aiming at reducing the color degradation introduced by CRT monitors. According to the first, a mathematical model of the CRT is developed and an approximation of its inverse is used to pre-process the images to be reproduced in such a way that color distortion is minimized. The second method is similar to the first one, the only difference consisting in the use of a neural network to model the behavior of the CRT. According to both strategies, training against a set of reference colors is needed to tune the parameters of the model. Experiments were carried out to evaluate the performance of the proposed methods. Upon analysis of the results, the superior accuracy of the neural- based system comes out, due to its capability of dealing with the non-linear, non-ideal behavior of commercial monitors. On the other hand, more flexibility of use can be achieved through a mathematical description of the CRT, which in some application scenarios makes such a solution a more desirable one.

In this paper, we developed a new method of the document analysis technique to extract the patient information directly from the scanned image. This information can link with Radiology Information System (RIS) or Hospital Information System, and the scanned images can then be filed into database automatically. The efficiency this method offers can greatly simplify the image filing process and improve the user friendliness of the overall image digitization system. Moreover, it solves the automatic information input problem in a very economical way. We believe the success of this technique will benefit the development of the PACS and teleradiology.

This paper describes a motion Wavelet transform Zero Tree (WZT) codec which achieves good compression ratios and can be implemented in a single ASIC of modest size (and very low cost). WZT includes a number of trade-offs which reduce the compression rate but which simplify the implementation and reduce the cost. The figure of merit in our codec is ASIC silicon area required, and we are willing to sacrifice some rate/distortion performance with respect to the best available algorithms to that goal. The codec employs a group of pictures (GOP) of two interlaced video frames (i.e., four video fields). Each such field is coded by the well-known transform method, whereby the image is subjected to a 2-D linear transform, the transform values are quantized, and the resulting values coded (e.g., by a zero-tree method). To that goal we are using 3D wavelet transform, dyadic quantization and various entropy codecs. In the temporal direction a temporal transform is used instead of motion estimation. Some of the technical innovations that enable the above features set are: (1) Edge filters which enable blockwise processing while preserving quadratic continuity across block boundaries, greatly reducing blocking artifacts. (2) Field image compression which reduces memory requirements for fields within a GOP.

This paper presents a high performance block-based wavelet image coder which is designed to be of very low implementational complexity yet with rich features. In this image coder, the Dual-Sliding Wavelet Transform (DSWT) is first applied to image data to generate wavelet coefficients in fixed-size blocks. Here, a block only consists of wavelet coefficients from a single subband. The coefficient blocks are directly coded with the Low Complexity Binary Description (LCBiD) coefficient coding algorithm. Each block is encoded using binary context-based bitplane coding. No parent-child correlation is exploited in the coding process. There is also no intermediate buffering needed in between DSWT and LCBiD. The compressed bit stream generated by the proposed coder is both SNR and resolution scalable, as well as highly resilient to transmission errors. Both DSWT and LCBiD process the data in blocks whose size is independent of the size of the original image. This gives more flexibility in the implementation. The codec has a very good coding performance even the block size is (16,16).

An integrated system for the robust coding of topological mesh data of arbitrary 3D graphic models is investigated in this work. The proposed system can achieve higher error resiliency with a low bit-rate overhead. This system mainly consists of two major modules, i.e. the segmentation module and the reversible variable length coding (RVLC) module. The segmentation module is used to divide an arbitrary 3D mesh into a group of smaller, uniform and independent segments depending on the error rate. Errors introduced in network transmission can be limited to the current segment instead of the whole mesh, which reduces the error propagation length drastically. The reversible variable length coding module is applied to each individual segment. It allows the recovery of a large portion of data from a corrupted segment due to the two-way decoding capability of RVLC. The amount of retransmitted data can thus be greatly reduced. In this research, two specific types of RVLC are considered, their parameters are carefully selected to match the symbol probability distributions. Experimental results show that an average overhead of 10 - 20% is required by the proposed scheme in comparison with the original error-free coding technique for the 300 testing 3D graphic models to given an excellent performance in the presence of noise.

A feedback-based Internet video transmission scheme based on the ITU-T H.263+ is presented. The proposed system is capable of continually adjusting the stream size to avoid the congestion in response to network condition changes. It consists of several major components such as TCP-friendly end- to-end congestion control with available bandwidth estimation, encoding frame rate control and transmission buffer smoothing at the server. These components are designed to meet the low computational complexity requirement so that the whole system can operate in real time. Among these, video-aware congestion control, which is called the receiver-based congestion control mechanism (RCCM), and the variable frame rate H.263+ encoding are the two key features. Through a seamless integration of these feature components, it is demonstrated that network adaptivity is enhanced to mitigate the packet loss and the bandwidth fluctuation, resulting in a smoother video experience at the receiver.

In this paper we present a novel synthesis of embedded compression and texture recognition. In order to maximize the information content of a transmitted image, a texture recognition and segmentation algorithm is used to identify potential areas of disinterest, and the compression algorithm regionally compresses the image, allocating fewer bits to textured regions. In this method, higher resolution is achieved in areas that potentially are of more interest to a viewer.

Video segmentation is an important first step towards automatic video indexing, retrieval, editing, and etc. However, the 'large' property of video makes it hard to handle in real time. To fulfill the goal of real-time processing, several factors need to be considered. First of all, indexing video directly in the compressed-domain offers the advantages of fast processing upon efficient storage. Secondly, extracting simple features with fast algorithms is no doubt helpful in speeding up the process. The questions are what kind of simple feature can characterize the changing statistics and what kind of algorithm can provide such feature with fast executability. In this paper, we propose a new automatic video segmentation scheme that utilizes wavelet transformation based on the following consideration: wavelet is a nice tool for subband decomposition, it encodes both frequency and spatial information; more over, it is easy to program and fast to execute. In the last decade or so, wavelet transform is emerged to image/video signal processing for analyzing functions at different levels of details. In particular, wavelet, as a tool, has been widely used in the area of image compression. In image compression, it is possible to recover a fairly accurate representation of the image by saving the few largest wavelet coefficients (and throwing away part or all of the smaller coefficients). By using this property, we extract a discrimination signature of each image from a few large coefficients for each color channel. The system works on the compressed video that does not require full decoding of the video and performs a wavelet transformation on the extracted video data. The signature (as feature) is extracted from the wavelet coefficients to characterize the changing statistics of shot transitions. Cuts, fades, and dissolve are detected based on the analysis of the changing statistics curve.

It is well known that non-uniformity noise in focal-plane array (FPA) sensors, which is due to the pixel-to-pixel variation in the detectors' responses, can considerably degrade the quality of images since it results in a fixed pattern that is superimposed on the true scene. This paper addresses the benefits of non-uniformity correction (NUC) pre- processing on the performance of a number of existing post- processing algorithms. The post-processing applications considered are image registration (motion compensation) and high-resolution image reconstruction from a sequence of blurred and undersampled images. The accuracy of motion estimation in the presence of non-uniformity noise is investigated in terms of the standard deviation of the gain and the offset of the FPA detectors. Two recently reported scene-based NUC techniques are employed to pre-process the FPA output prior post-processing. The first NUC technique is based on the assumption that all pixels experience the same range of irradiance over a sufficiently long sequence of frames (constant-range assumption), and the second NUC technique is registration-based and can be used recursively. It is shown that NUC improves the accuracy of registration, and consequently, enhances the benefit of high-resolution reconstruction. It is also shown that under cases of severe non-uniformity noise, a recursive application of the registration-based NUC may further improve the accuracy of registration and high-resolution image reconstruction. The results are illustrated using real and simulated data.

The problem addressed in this paper is that of clustering image pixels into regions of homogenous geological texture. Future rovers on Mars will need to be able to intelligently select data collection targets. One goal of intelligent data selection for maximizing scientific return is to sample all distinct types of rocks that may be encountered. Different rock types often have a characteristic visual texture, thus visual texture is rich source of information for separating rocks into different types. Recent work on using texture to segment images has been very successful on images with homogenous textures such as mosaics of Brodatz textures and some natural scenes. The geologic history of a rock leads to irregular shapes and surface textures. As a result, the textures in our images are not as homogeneous as those in Brodatz mosaics. Our approach is to extract textural information by applying a bank of Gabor filters to the image. The resulting texture vectors are then clustered. Banks of filters constrain the relationships of the filter parameters both within a single filter and between filters. Often researchers have used parameter values that are thought to correspond to the human visual system, however the effects of adjusting these parameters have not been thoroughly studied. We systematically explore tradeoffs in the parameter space of the filter bank and quantify the effects of the takeoffs on the quality of the resulting clusters.

The potential of high-resolution radar and optical imagery for synoptic and timely mapping in many applications is well- known. Numerous methods have been developed to process and quantify useful information from remotely sensed images. Most image processing techniques use texture based statistics combined with spatial filtering to separate target classes or to infer geophysical parameters from pixel radiometric intensities. The use of spatial statistics to enhance the information content of images, thereby providing better characterization of the underlying geophysical phenomena, is a relatively new technique in image processing. We are currently exploring the relationship between spatial statistical parameters of various geophysical phenomena and those of the remotely sensed image by way of principal component analysis (PCA) of radar and optical images. Issues being explored are the effects of noise in multisensor imagery using PCA for land cover classifications. The differences in additive and multiplicative noise must be accounted for before using PCA on multisensor data. Preliminary results describing the performance of PCA in the presence of simulated noise applied to Landsat Thematic Mapper (TM) images are presented.

The technique of integral projections is used to perform co- registration of data from a wedge spectrometer instrument that has been developed by NASA Goddard Space Flight Center. The spectrometer is currently being flown on a plane and operates in the 1 - 2.5 micron range. The technique involves a number of steps. First, an algorithm was developed to calculate the absorption bands that occur within the spectral region. At this point the method of Integral Projections is used to vectorize the image. The Integral Projections method performs a number of key functions in the registration process: increases SNR, reduces affects of spatial non-uniformities within the data, and results in a much faster algorithm since the operation is on vectors. The final step is to register the zero crossing of the second derivative of the vectors. Two issues encountered with co-registration is dealing with the absorption bands that occur within the spectral region of interest and the multiple problem of recognizing features that are not only shifting in x and y but also appear different at different wavelengths. Results will be presented in which the application of our algorithm obtains the appropriate x,y shifts necessary to reconstruct a registered data cube.

In this work a computational approach for detecting and quantifying diabetic retinopathy is proposed. Particular attention has been paid to the study of Foveal Avascular Zone (FAZ). In fact, retinal capillary occlusion produces a FAZ enlargement. Moreover, the FAZ is characterized by qualitative changes showing an irregular contour with notchings and indentations. Our study is mainly focused on the analysis of the FAZ and on the extraction of a proper set of features to quantify FAZ alterations in diabetic patients. We propose an automatic segmentation procedure to correctly identify the FAZ boundary. The method was derived from the theory of active contours, also known as snakes, along with genetic optimization. Then we tried to extract features which can capture not only the size of the object, but also its shape and spatial orientation. The theory of moments provides an interesting and useful way for representing the shape of objects. We used a set of region and boundary moments to obtain a FAZ description which is complete enough for diagnostic purposes and in order to assess the effectiveness of moment descriptors we performed several classification experiments to discriminate diabetic from non-diabetic subjects. We used a neural network-based classifier, optimized for the problem, which is able to perform feature selection at the same time as learning, in order to extract a subset of features. The theory of moments provided us with an interesting and useful tool for representing the shape characteristics. In this way we were able to transform the qualitative description of the FAZ used by ophthalmologists into quantitative measurements.

In many applications such as in airborne and terrestrial reconnaissance, robotics, medical imaging, and machine vision systems, the images of a video sequence are severely distorted by vibrations. Superresolution algorithms are suitable for restoring an image from a low-frequency vibrated sequence because of high correlation between the frames and inherent interframe motion. However, we show in this work that superresolution algorithms which were developed for general types of blurs should be adapted to the specific characteristics of low-frequency vibration blur. We demonstrate that in the case of image sequences distorted by vibration, the images should be selected prior to processing. We found empirical selection criteria and propose a selection procedure.

In the typical stereo vision system, when the fixation points of the left and right images are mismatched or the moving object is not in the center of the screen, not only the observer is fatigued and unconscious of three-dimensional effect, but also hard to track the moving object. Therefore, stereo object tracking is necessary that controls the convergence angle, the angle between the optic axes of cameras and locates the moving object at the center of the screen. In this paper, we propose and demonstrate a stereo object tracking simulator based on the optical JTC.

Using the lifting step approach for wavelet decomposition, Sweldens has recently introduced a fully integer based filtering method. There are several advantages to such an approach, one of the most interesting is the possibility to use wavelets for efficient lossless coding. However, this scheme is also interesting in case of lossy compression, especially for 'real-time' or 'low-cost' applications. In the PC based world, integer operations are more efficient than their floating-point counterparts, allowing much faster processing. In case of hardware implementations, integer based arithmetic units are much cheaper than those capable of handling floating points. In terms of memory usage, integer decomposition reduces the demands on the system by at least a factor two. For these reasons, we are interested in considering integer based filtering for lossy image compression as well. This raises an important question: what additional losses, if any, occur when using integer based wavelet decompositions in place of the usual floating point approach? First we compare the compressed images using standard SNR and other simple metrics. Next we evaluate our results using visually weighted objective metrics. This allows us to fully evaluate integer wavelet decomposition when applied to lossy image compression across a range of bit rates, filter characteristics and image types.

EBCOT is an efficient image coding technique, which divides each subband into independently coded blocks. It is therefore inherently more error resilient than many other wavelet-based schemes. However, the loss of data of a block in any lower frequency subband in EBCOT can still degrade the perceptual image quality considerably. This paper discusses the use of reversible variable length codes (RVLC) and data partitioning for coding lower frequency subbands in EBCOT, instead of the use of arithmetic codes. RVLCs are known to have superior error recovery properties due to their two-way decode capability. It is demonstrated that the proposed approach has very little additional bit-rate overhead and significantly improved the performance in the presence of errors.

In this paper, we propose a novel object driven, block based algorithm for the compression of stereo image pairs. The algorithm effectively combines the simplicity and adaptability of the existing block based stereo image compression techniques with an edge/contour based object extraction technique to determine appropriate compression strategy for various areas of the right image. Extensive experiments carried out support that significant improvements of up to 20% in compression ratio can be achieved by the proposed algorithm, compared with the existing stereo image compression techniques. Yet the reconstructed image quality is maintained at an equivalent level in terms of PSNR values. In terms of visual quality, the right image reconstructed by the proposed algorithm does not incur any noticeable effect compared with the outputs of the best algorithms. The proposed algorithm performs object extraction and matching between the reconstructed left frame and the original right frame to identify those objects that match but are displaced by varying amounts due to binocular parallax. Different coding strategies are then applied separately to internal areas and the bounding areas for each identified object. Based on the mean squared matching error of the internal blocks and a selected threshold, a decision is made whether or not to encode the predictive errors inside these objects. The output bit stream includes entropy coding of object disparity, block disparity and possibly some errors, which fail to meet the threshold requirement in the proposed algorithm.

Transform coding has been the focus of the research for image compression. In previous research, the Amplitude and Group Partitioning (AGP) coding scheme is proved to be a low complexity algorithm with higher performance, clearly one of the state-of-art transform coding techniques. However, the previous AGP is used along with the Discrete Cosine Transform (DCT) and the discrete wavelet transform. In this paper, a different transform, the Lapped Orthogonal Transform (LOT), replaces the DCT in conjunction with the AGP. This is the first time LOT and AGP have been combined in a coding method. The definition and design of the LOT are discussed. An objective metric to measure the performance of transform, coding gain, is calculated for both the DCT and the LOT. The LOT has slightly higher coding gain than the DCT. The principles of the LOT based AGP image codec (LOT-AGP) are presented and a complete codec, encoder and decoder, is implemented in software. The performance of the LOT-AGP is compared with other block transform coding schemes: the baseline JPEG codec and the DCT based AGP image codec (DCT- AGP) by both objective evaluation and subjective evaluation. The Peak Signal to Noise Ratio (PSNR) is calculated for these three coding schemes. The two AGP codecs are much better than the JPEG codec on PSNR, from about 1.7 dB to 3 dB depending on bit rate. The two AGP schemes have PSNR differences only to a small degree. Visually, the LOT-AGP has the best-reconstructed images among these three at all bit rates. In addition, the coding results of two other state-of-art progressive image codecs are cited for further comparison. One is the Set Partitioning in Hierarchical Trees (SPIHT) algorithm with a dyadic wavelet transform, and the other is Tran and Nguyen's method with the generalized LOT transform. The AGP coding and the adaptive Huffman entropy coding of LOT-AGP are less complex and the memory usage is smaller than in these two progressive codecs. Comparing these three codecs, i.e. the LOT-AGP and the two progressive codecs in PSNR small only small differences in PSNR. SPIHT has about 1 dB higher PSNR than the LOT-AGP and Tran and Nguyen's method for the test image Lena. For the test image Barbara, the PSNR of the LOT- AGP is about 0.5 dB higher than that of the SPIHT and 0.5 dB lower than that of Tran and Nguyen's method. This low- complexity and high performance codec may provide a new direction for the implementation of image compression.

This paper focuses on lossless medical image compression methods for 3D volumetric medical images that operate on three-dimensional (3D) reversible integer wavelet transforms. We offer an application of the Set Partitioning in Hierarchical Trees (SPIHT) algorithm to volumetric medical images, using a 3D wavelet decomposition and a 3D spatial dependence tree. The wavelet decomposition is accomplished with integer wavelet filters implemented with the lifting method, where careful scaling and truncations keep the integer precision small and the transform unitary. We have tested our encoder on volumetric medical images using different integer filters and different coding unit sizes. The coding unit sizes of 16 and 8 slices save considerable memory and coding delay from full sequence coding units used in previous works. Results show that, even with these small coding units, our algorithm with certain filters performs as well and sometimes better in lossless coding than previously coding systems using 3D integer wavelet transforms on volumetric medical images.

In this research, we propose an approach to combine the image compression and the image watermarking schemes in an effective way. The image coding scheme under our consideration is EBCOT (Embedded Block Coding with Optimized Truncation) which is the basis of JPEG2000 VM (Verification Model). The watermark is embedded when the compressed bit-stream is formed, and can be detected on the fly during image decompression. With the proposed integrated method, watermark embedding and retrieval processes can be done very efficiently compared with existing watermarking schemes. The embedded watermark is robust against various signal processing attacks including compression and filtering while the resulting watermarked image maintains good perceptual quality. Furthermore, the watermark can be detected progressively and ROI (Region of Interest)-based watermarking can be easily accomplished.

When multispectral images are being losslessly compressed, if the inter-band correlation of the data is weak, as it usually occurs for data with few and sparse spectral bands, a 3D prediction may lead to negligible coding benefits. In this case, advantage may be taken from a bidirectional spectral prediction, in which once the (k - 1)^st band is available, first the k^th band is skipped and the (k + 1)^st band is predicted from the (k - 1)^st one; then, both these two bands are used to predict the k^th band in a spatially causal but spectrally noncausal fashion. Starting from an extremely sophisticated and effective scheme for multispectral prediction based on fuzzy-logic concepts, the causal and noncausal 3D prediction strategies are compared and discussed varying with the spectral correlations of the data. Experiments on Landsat TM data show that a certain gain in bit rate can be obtained at no additional cost, by simply devising a scan order in which some bands are preliminarily skipped and then bidirectionally predicted.

Few studies have reported on multi-spectral color calibration using statistical information from low altitude images as the reference for colors in remote sensing image processing. For multi-spectral color calibration, a new statistical adjustment technique was applied based on the fact that low altitude images can offer relatively more accurate colors than high altitude images, and the information from the two comparable images has similar characteristics in a statistical sense. Therefore, by decomposing colors from both lower and higher altitude image data, and matching histograms for each corresponding color component, the colors of higher altitude images can be calibrated. With a heuristic measurement method, the resultant images had more than 80% similarity to the reference image, whereas, with the relative difference method, the resultant images were more than 90% closer to the reference image.

The progress of robotic and machine vision technology has increased the demand for sophisticated methods for performing automatic harvesting of fruit. The harvesting of fruit, until recently, has been performed manually and is quite labor intensive. An automatic robot harvesting system that uses machine vision to locate and extract the fruit would free the agricultural industry from the ups and downs of the labor market. The environment in which robotic fruit harvesters must work presents many challenges due to the inherent variability from one location to the next. This paper takes a step towards this goal by outlining a machine vision algorithm that detects and accurately locates oranges from a color image of an orange tree. Previous work in this area has focused on differentiating the orange regions from the rest of the picture and not locating the actual oranges themselves. Failure to locate the oranges, however, leads to a reduced number of successful pick attempts. This paper presents a new approach for orange region segmentation in which the circumference of the individual oranges as well as partially occluded oranges are located. Accurately defining the circumference of each orange allows a robotic harvester to cut the stem of the orange by either scanning the top of the orange with a laser or by directing a robotic arm towards the stem to automatically cut it. A modified version of the K- means algorithm is used to initially segment the oranges from the canopy of the orange tree. Morphological processing is then used to locate occluded oranges and an iterative circle finding algorithm is used to define the circumference of the segmented oranges.

Morphological image processing has been widely used to process binary and grayscale images, with morphological techniques being applied to noise reduction, image enhancement, and feature detection. Relying on an ordering of the data, morphology modifies the geometrical aspects of an image. Extending morphological operators to color image processing has been problematic because it is not easy to define geometry of a vector-valued function and ordering of vectors is not straightforward. We describe and demonstrate color morphological operators based on a combination of reduced- ordering and conditional-ordering of the underlying data.

Image enhancement has, in general, a substantial effect on human detection performance of targets embedded in infrared images. However, different enhancement techniques, as well as different surrounding conditions, cause this effect to vary. In this paper, we analyze the effect of several histogram modeling enhancement methods on human detection performance, according to specially designed psychophysical experiment results. We also present, using the Analysis of Variance (ANOVA) statistical tool, how this effect varies according to different surrounding conditions, and whether the effect is statistically significant. A quantitative empirical criterion for enhancement efficiency is introduced and used to show how histogram modeling enhancement can considerably improve poor detection performance associated with natural images, provided that the applied enhancement method does not change drastically the natural image Pixel Distribution Function (PDF). On the other hand, if the criterion is not achieved, histogram modeling enhancement has negligible influence.

To support fast browsing of large scale images in a client/server environment, we propose a prediction-based server prefetching and client caching algorithm. By prefetching the predicted data using available network resources, the system can effectively reduce the size of data needed for the next access. While traditional browsing cache systems can not address both nonstationary and stationary browsing behaviors at the same time, our G/L-cache system is adaptive to dynamic user accesses. As a result, the response time for an interactive browsing system can be greatly reduced.

In this paper we propose a method for detecting moving objects in image sequence observed from a moving platform using optical flow. This problem is difficult because moving observer (i.e. camera) causes apparent motion in the image even for stationary environment. The method can be applied to many situations, such as a robot vision and an obstacle detection for an autonomous vehicle system. We assume that observer motion parameter (translation and rotation) is known and image system is modeled by perspective projection. For the problem, some methods have been proposed, in which the complex logarithm mapping, the estimation of Focus of Expansion and the depth of objects are used. For a given motion parameter of camera, we can formulate motion field constraint (MFC) in the image plane which is satisfied by the relative movement of stationary environment against camera motion. On the other hand, the motion vector in the image plane, which is called motion field, is estimated by the well-known optical flow constraint (OFC). Our main idea is to use the difference between two estimation results. One is the solution of minimizing least squared OFC subjected with MFC, and the other is the solution of that without MFC. For the stationary environment region, the difference between two is small and the difference tends to be large at the moving region. Therefore, the suitable criterion for these values will separate two regions precisely. In our study, two criteria are proposed and are investigated. One criterion uses squared residual of OFC with and without MFC. Another criterion uses directional error between two solutions. The validity of our method is shown through some examples, and the obtained results show the latter criterion gives more accurate estimation than the former one.

The removal of noise in image is one of the current important issues. It is also useful as a preprocessing for edge detection, motion estimation and so on. In this paper, an adaptive weighted median filter utilizing impulsive noise detection is proposed for the removal of impulsive noise in digital images. The aim of our proposed method is to eliminate impulsive noise effectively preserving original fine detail in images. This aim is same for another median-type nonlinear filters try to realized. In our method, we use weighted median filter whose weights should be determined by balancing between the signal preserving ability and noise reduction performance. The trade off between these two inconsistent properties is realized using the noise detection mechanism and optimized adaptation process. In the previous work, threshold value between the signal and the output of the median filter have to be decided for the noise detection. Adaptive algorithm for optimizing WM filters uses the teacher image for training process. In our method, following two new approaches are introduced in the filtering. (1) The noise detection process uses the discriminant method to the histogram distribution of the derivation from median filter output. (2) Filter weights which have been learned by uncorrupted pixels and their neighborhood without the original image are used for the restoration filtering for noise corrupted pixels. The validity of the proposed method is shown through some experimental results.

This paper presents a new algorithm for object motion detection and trajectory tracking. This method was developed as part of a machine vision system for human fertility analysis. Fertility analysis is based on the amount of spermatozoa in semen samples and their type of movement. Two approaches were tested to detect the movement of the spermatozoa, image subtraction, and optical flow. Image subtraction is a simple and fast method but it has some complications to detect individual motion when large amounts of objects are presented. The optical flow method is able to detect motion but it turns to be computationally time expensive. It does not generate a specific trajectory of each spermatozoon, and it does not detect static spermatozoa. The algorithm developed detects object motion through an orthogonal search of blocks in consecutive frames. Matching of two blocks in consecutive frames is defined by square differences. A dynamic control array is used to store the trajectory of each spermatozoon, and to deal with all the different situations in the trajectories like, new spermatozoa entering in a frame, spermatozoa leaving the frame, and spermatozoa collision. The algorithm developed turns out to be faster than the optical flow algorithm and solves the problem of the image subtraction method. It also detects static spermatozoa, and generates a motion vector for each spermatozoon that describes their trajectory.

A novel algorithm based on the sigma filter for processing multicomponent images is proposed. The noise suppression ability of the proposed vector filtering algorithm is better than, e.g., that of the standard sigma filter. Moreover, the added modifications make the filter able to remove impulsive noise. The proposed vector filter takes into account the mutual correlation between image components and preserves object edges and fine details even when the contrasts of the component images of multichannel data are low. The comparative analysis of filter performance is done both visually and using several quantitative criteria. Both simulated and real color and multichannel radar images are studied. It is shown that the modified vector sigma filter outperforms many component and vector filters. Two modifications are considered -- for cases of additive and multiplicative noise. Examples of the filter performance for processing real images formed by multipolarization/multifrequency side-look aperture radars are presented.

Several engineering applications are concerned with the accurate and efficient identification of the least-squares (LS) solution. The computational and storage requirements to determine the LS solution become prohibitively large as the dimensions of the problem grow. This paper develops an algorithm which receives the least squares solution based on a steepest descent formulation. Among the advantages of this approach are improvements in computational and resource management, and ease of hardware implementation. The gradient matrix is evaluated using 2-D linear convolutions and an in- place update strategy. An iterative procedure is outlined and the regularized and unregularized LS solutions can be recovered. The extent of regularization is suitably controlled and imposes some constraints on the step size for steepest descent. The proposed approach is examined in the context of digital image restoration from spatially invariant linear blur degradation and compared with alternate strategies performing LS recovery.

Gridding reconstruction is a method to derive data on a Cartesian grid from a set of non-uniformly sampled measurements. This method is appreciated for being robust and computationally fast. However, it lacks solid analysis and design tools to quantify or minimize the reconstruction error. Least squares reconstruction, on the other hand, is another method which is optimal in the sense that it minimizes the reconstruction error. This method is computationally intensive and, in many cases, sensitive to measurement noise; hence it is rarely used in practice. Despite the seemingly different approaches of reconstruction, the gridding and least squares reconstruction methods are shown to be closely related. The similarity between these two methods is accentuated when they are properly expressed in a common matrix form. It is shown that the gridding algorithm can be considered an approximation to the least squares method. The optimal gridding parameters are defined as ones yielding the least approximation error. These parameters are calculated by minimizing the norm of an approximation error matrix. This method is used to find the optimal density compensation factors which minimize the weighted approximation error. An iterative method is also proposed for joint optimization of the interpolating kernel and the deapodization function. Some applications in magnetic resonance imaging are presented.

It is well known that using different versions of a scene perturbed with different blurs improves the quality of the restored image compared to using a single blurred image. In this paper, we consider images perturbed with large defocus blurs. In the case where two different blurs are used, we characterize the influence of the relative diameter of both blurs on the restoration quality. We show that using two different blurs significantly increases the robustness of the restoration to a mismatch between the kernels used in designing the restoration algorithm and the actual blurs that have perturbed the image. We finally show that using three different kernels may not improve the restoration performance compared to the two-kernel approach, but still improves the robustness to kernel estimation.

Properties and the structure of an image can be studied through its various connected components. In this paper, we apply a theory of parameter-dependent connected components developed by us in a previous work to analyze the structure of an images.

This research investigates the use of image content analysis techniques as a tool of understanding image content, organizing image databases and choosing appropriate low level features as well as semantic meanings for image indexing and retrieval. An intelligent image indexing and query system consisting of semantic classification, composite indexing and interactive query is proposed. In this system, a large collection of images with great varieties is analyzed by the content and categorized into different classes according to distinct characteristics. The semantics of feature descriptors and the relationship between feature descriptors and image contents are then explored. Finally, a composite indexing and interactive retrieval procedure using best low-level features and high-level understanding is developed to achieve a robust image query performance.

The ability to extract useful and robust features from sensor data of vehicles in moving traffic is highly dependent on a number of factors. For imaging sensors that produce a two- dimensional representation of an observed scene such as a visible light camera, the principal factors influencing the quality of the acquired data include the ambient lighting and weather conditions as well the physical characteristics of the vehicles whose images are captured. Considerable variability in the ambient lighting conditions in combination with material characteristics may cause radically different appearances for various surfaces of a vehicle when viewed in the visible wavelengths. Infrared sensors, on the other hand, produce images that are far less sensitive to variations in ambient lighting conditions, but may not provide sufficient information that can be used to discriminate among vehicles. Combining information from these sensors provides the basis for exploiting the relative strengths of each sensor domain while attenuating the weaknesses that exist in single systems. This paper presents a basic framework for combining information from multiple sensor systems by describing methodologies for geometrically transforming between image spaces and extracting features using a multi-dimensional approach that exploits information gathered at different wavelengths. The potential use of point sensors (such as acoustic and microwave detectors) in combination with imaging sensors is also discussed.

A method for reconstruction and restoration of super resolution images from sequences of noisy low-resolution images is presented. After estimating the projective transformation parameters between a selected reference image and the observed degraded image sequence frames, the data is rearranged into a sequence with only quantized sub pixel translations. Next, the imaging system's point spread function (PSF) and the auto-correlation function of the image are estimated with a resolution higher than that of the super resolution image. The coefficients of the FIR Wiener filter are computed, low-pass filtered, and decimated so a polyphase filter bank is obtained. Each one of the images in the translated rearranged sequence is filtered by its corresponding polyphase filter. These filtering results are summed and locally normalized according to the apparent data. The super resolution result is refined by estimating the values of pixels that could not be reconstructed by interpolation. The use of the polyphase filters allows exploitation of the input data without any averaging operations needed when implementing conventional FIR Wiener filtering. The presented experimental results show good resolution improvement in presence of noise.

This paper describes the design of a document region classifier. The regions of a document are classified as large text regions, LTR, and non-LTR. The foundations of the classifier are derived from human visual perception theories. The theories analyzed are texture discrimination based on textons, and perceptual grouping. Based on these theories, the classification task is stated as a texture discrimination problem and is implemented as a preattentive process. Once the foundations of the classifier are defined, engineering techniques are developed to extract features for deciding the class of information contained in the regions. The feature derived from the human visual perception theories is a measurement of periodicity of the blobs of the text regions. This feature is used to design a statistical classifier based on the minimum probability of error criterion to perform the classification of LTR and non-LTR. The method is test on free format low resolution document images achieving 93% of correct recognition.

In this paper we extend the autoregressive (AR) model to the multilevel AR model with wavelet transformation, in order to get the AR coefficients at each level as a set of shape descriptors for every level. To get the multilevel AR model, we use the wavelet transformation such as Haar wavelet to a boundary data. Then real AR and complex-AR (CAR) models are adopted to the multilevel boundary data of a shape to extract the features at each level. Furthermore we present the relation of the autocorrelation coefficients between adjacent resolution levels to elucidate the relation between AR model and wavelet transformation. Some experiments are also shown for the multilevel AR and CAR models with a certain similarity measure.

The radar backscatter from the ocean surface, called sea clutter, is modulated by the surface wave field. A method was developed to estimate the near-surface current, the water depth and calibrated surface wave spectra from nautical radar image sequences. The algorithm is based on the three- dimensional Fast Fourier Transformation (FFT) of the spatio- temporal sea clutter pattern in the wavenumber-frequency domain. The dispersion relation is used to define a filter to separate the spectral signal of the imaged waves from the background noise component caused by speckle noise. The signal-to-noise ratio (SNR) contains information about the significant wave height. The method has been proved to be reliable for the analysis of homogeneous water surfaces in offshore installations. Radar images are inhomogeneous because of the dependency of the image transfer function (ITF) on the azimuth angle between the wave propagation and the antenna viewing direction. The inhomogeneity of radar imaging is analyzed using image sequences of a homogeneous deep-water surface sampled by a ship-borne radar. Changing water depths in shallow-water regions induce horizontal gradients of the tidal current. Wave refraction occurs due to the spatial variability of the current and water depth. These areas cannot be investigated with the standard method. A new method, based on local wavenumber estimation with the multiple-signal classification (MUSIC) algorithm, is outlined. The MUSIC algorithm provides superior wavenumber resolution on local spatial scales. First results, retrieved from a radar image sequence taken from an installation at a coastal site, are presented.

Three-step search (TSS) has been studied for low bit-rate video communications because of significantly reduced computation, simplicity and reasonable performance in the motion estimation. Many other modified TSS algorithms have been developed for higher speedup of computation and improved error performance of motion compensation. Among the modified TSS algorithms, new three-step search (NTSS) shows very good error performance with more reduced computation on the average. However, the method can exceed about 30% compared with the computation of TSS for some cases. It can be serious problem in real-time video coding at the worst case. This paper proposes an efficient and adaptive three-step search (EATSS) algorithm with adaptive search strategy considering unequal subsampling and partial distortion elimination (PDE). Proposing search strategy reduces useless checking points of the first step in the NTSS by using initial sum of absolute difference (SAD) and predefined threshold of SAD. Instead of checking 17 candidate points in the first step as the NTSS, our search algorithm starts with 1 or 9 checking points according to the comparison between initial SAD and predefined threshold of SAD. Experimentally, our algorithm shows good performance in terms of PSNR of predicted images and average checking points for each block compared with NTSS and TSS.

Singular points are one of the characteristic features of fingerprint images. In this paper, a new way of finding these singular points from the directional image is proposed. The image used in this system is a 256 X 256 rectangular image from which the hexagonal image will be extracted. After producing the directional image, two other images will be constructed for the core & delta candidate points. These pair of points will be selected using a method based on Hough transform and local maximum algorithm. Based on the position and number of these points a one-byte code will be assigned to that class. By performing the Fourier image of the original picture, an eight-byte code will be assigned to the image using a 4 X 4 wedge-ring detector applied to the transformed image. These codes will be used to show the dissimilarity among similar images.

Image and Video indexing is becoming popular with the increasing volumes of visual information that is being stored and transmitted in various multimedia applications. An important focus of the upcoming MPEG 7 standard is on indexing and retrieval of multimedia data. The visual information can be indexed using the spatial (color, texture, shape, sketch, etc.) and temporal (motion, camera operations, etc.) features. Since multimedia data is likely to be stored in compressed form, indexing the information in compressed domain entails savings in compute time and storage space. In this paper, we present a novel indexing and retrieval technique using vector quantization of color images. Color is an important feature for indexing the visual information. Several color based indexing schemes have been reported in the recent literature. Vector Quantization (VQ) is a popular compression technique for low-power applications. Indexing the visual information based on VQ features such as luminance codebook and labels have also been recently presented in the literature. Previous VQ-based indexing techniques describes the entire image content by modeling the histogram of the image without taking into account the location of colors, which may result in unsatisfactory retrieval. We propose to incorporate spatial information in the content representation in VQ-compressed domain. We employ the luminance and chrominance codebooks trained and generated from wavelet-vector-quantized (WVQ) images, in which the images are first decomposed using wavelet transform followed by vector quantization of the transform coefficients. The labels, and the usage maps corresponding to the utilization pattern of codebooks for the individual images serve as indices to the associated color information contained in the images. Hence, the VQ compression parameters serve the purpose of indexing resulting in joint compression and indexing of the color information. Our simulations indicate superior indexing and retrieval performance at a reduced computational complexity. Performance analysis over a variety of color images using the proposed technique and other competing techniques in the literature is presented in the paper.

Conventional mean squared error based methods for objective image quality assessment are not well correlated with human evaluation. The design of better objective measures of quality has attracted a lot of attention and several image quality metrics based explicitly on the properties of the Human Visual System (HVS) have been proposed in recent years. However, only in a few cases has the performance of such metrics been demonstrated on real images. In accounting for visual masking, all of these metrics assume that the multiple channels mediating visual perception are independent of each other. Recent neuroscience findings and psychophysical experiments have established that there is interaction across the channels and that such interactions are important for visual masking. In this work, we propose the Picture Distortion Metric (PDM) which integrates these new visual masking properties, and we evaluate its performance for image coding applications. We evaluate the performance at medium to high range of quality to predict subjective scores on a 0 - 10 numerical scale, and on a wide range of quality for the 1 - 5 CCIR impairment scale.

A new smart interactive video object generation (SIVOG) system targeting at general semantic video object segmentation at pixel-wise accuracy is proposed. SIVOG identifies and extracts the semantic video object from image sequence with user interaction. The system consists of several basic components: semantic level user interaction, smart-processing kernel, object tracking, boundary update and error correction. It allows user to enter the semantic information with the least amount of mouse clicking and movement. Then, the user-input semantic information will be analyzed and interpreted in terms of low-lever features. Finally, the user can correct erroneous regions in the segmentation process. The proposed system is evaluated for several typical MPEG-4 test sequences.

A new blind inverse halftoning algorithm based on a nonlinear filtering technique of low computational complexity and low memory requirement is proposed in this research. It is called blind since we do not require the knowledge of the halftone kernel. The proposed scheme performs nonlinear filtering in conjunction with edge enhancement to improve the quality of an inverse halftoned image. Distinct features of the proposed approach include: efficiently smoothing halftone patterns in large homogeneous areas, additional edge enhancement capability to recover the edge quality and an excellent PSNR performance with only local integer operations and a small memory buffer.

A new method recently developed for restoration of motion blurred images is investigated and implemented for the special complicated case of image blur due to sinusoidal vibrations. Sinusoidal vibrations are analyzed here in the context of blur identification and image restoration. The main novel achievement of this work is the use of only a single vibrated blurred image as given information prior to the restoration process. The various cases of blur types that depend on the imaging conditions are considered in this paper.

A proposal for a model of the primary visual cortex is reported. It is structured with the basis of a simple unit cell able to perform fourteen pairs of different boolean functions corresponding to the two possible inputs. As a first step, a model of the retina is presented. Different types of responses, according to the different possibilities of interconnecting the building blocks, have been obtained. These responses constitute the basis for an initial configuration of the mammalian primary visual cortex. Some qualitative functions, as symmetry or size of an optical input, have been obtained. A proposal to extend this model to some higher functions, concludes the paper.

Extraction of 'man-made object' from stereoscopic aerial images is a problem which is not entirely solved. In this paper we propose an interactive approach to extract such objects. The disparity image is used to roughly delineate buildings. Then, the buildings are extracted according to their shape. An operator clicks inside the roof of the building of interest and chooses between two different algorithms, the rest of the process is entirely automatic. A very robust algorithm based on the Hough transform is used to extract rectangular buildings, whereas, a large spectrum algorithm based on snakes is used to extract more complex shapes. Our method gives good results in extracting buildings on IGN aerial images already rectified in epipolar geometry.

In this paper, we present a phase difference-based disparity estimation algorithm embedded in a cooperative multiscale scheme. We show that a reformulation of both the phase difference and the instantaneous frequency improves the density of the estimates. We define admissibility criteria of disparities. These criteria are based on both the magnitude and the phase linearity of the responses to increase the accuracy of the estimates but to the detriment of the density. By combining the multiscale estimates, the density is improved. Experimentations are reported for each criterion as well as for the multiscale combination. A performance analysis shows that the results obtained are very satisfactory.

In this paper, we present a system, called MediABS, for extracting key frames in a video segment. First we will describe the overall architecture of the system and we will show how our system can handle multiple video formats with a single video-processing module. Then we will present a new algorithm, based on color histograms. The algorithm exploits the temporal characteristic of the visual information and provides techniques for avoiding false cuts and eliminating the possibility of missing true cuts. A discussion, along with some results, will be provided to show the merits of our algorithm compared to existing related algorithms. Finally we will discuss the performance (in terms of processing time and accuracy) obtained by our system in extracting the key frames from a video segment. This work is part of the Mobile Agents Alliance project involving University of Ottawa, National Research Council (NRC) and Mitel Corporation.

We demonstrate the use of informational redundancy of signals for reducing the computational costs of signal processing. An accelerated algorithm for computing local, or short time, histograms of signals is proposed to support the idea of purposive use of signal redundancy for saving computational costs. Different rank-order filters using a large moving window can be designed on the base of the proposed algorithm. Computer tests and application of the algorithm to rank-order processing of real-life images are provided and discussed.

This paper reports on how multiple energy techniques in X and gamma-ray CT scanning are able to provide good results with the use of Statistical Pattern Classification theory. We obtained a set of four images with different energies (40, 60, 85 and 662 keV) containing aluminum, phosphorus, calcium, water and plexiglass, with a minitomograph scanner for soil science. We analyzed those images through both a supervised classifier based on the maximum-likelihood criterion under the multivariate Gaussian model and a supervised contextual classifier based on the ICM (iterated conditional modes) algorithm using an a priori Potts-Strauss model. A comparison between them was performed through the statistical kappa coefficient. A feature selection procedure using the Jeffries- Matusita (J-M) Distance was also performed. Both the classification and the feature selection procedures were found to be in agreement with the predicted discrimination given by the separation of the linear attenuation coefficient curves for different materials.

The technique of license plate recognition is important for traffic management's intelligentizing. A novel real time opto- electronic hybrid system is proposed and constructed for license plate recognition in this paper. A software algorithm accomplishes the extraction of characters from a license plate image. An optical coprocessor of correlation identification based on volume holographic storage and wavelet transform is developed to recognize the characters. Wavelet transform is introduced to improve the recognition accuracy. Its comparison with the conventional correlation is studied. Because the system can make a multichannel processing instantly in parallel, it is faster and more accurate than other systems for the license plate recognition. The feasibility of the system is testified by experimental results.

A new algorithm to compute a precise 3-D shape of a moving object for color motion stereo is described. Input data is obtained from a single color CCD camera and a moving belt. Three-dimensional shape recovery in motion stereo is formulated as a matching optimization problem of multiple color stereo images. It is shown that the problem of matching among multiple color motion stereo images can be carried out with use circular decorrelation of a color signal. Three- dimensional shape recovery using real color motion stereo images demonstrates a good performance of the algorithm in terms of reconstruction accuracy.

This paper describes the development of an optical mark reader that can be used for counting the examination score from the multiple-choice answer sheet. The system is developed based on PC-type microcomputer connecting to an image scanner. The system operations can be distinguished into two modes: learning mode and operation mode. In the learning mode, the model corresponding to each type of answer sheet is constructed by extracting all significant horizontal and vertical lines in the blank-sheet image. Then, every possibly cross-line will be located to form rectangular area. In the operation mode, each sheet fed into the system has to be identified by matching the horizontal lines detected with every model. The data extraction from each area can be performed based on the horizontal and vertical projections of the histogram. For the answer checking purpose, the number of black pixels in each answer block is counted, and the difference of those numbers between the input and its corresponding model is used as decision criterion. Finally, the database containing a list of subjects, students, and scores can be created. The experimental results on many styles of answer sheets show the effectiveness of such a system.

In this paper, the K-L expansion for feature extraction has been combined with an incoherent optical correlator, which was previously constructed for human face recognition. In this new approach, the eigenfaces are used as the image filters in the reference plane of the correlator. Since the face images can be approximated by different linear combinations of a relatively few eigenfaces, they can be efficiently distinguished from one another by a small set of the weight coefficients, which is derived by projecting the input image onto every eigenface. The optical correlator is used as the feature extractor and the optical correlation results between the input image and the eigenfaces are used as the features. As a result, the recognition features can be got at a relatively high speed. Because the face images in the training set are selected to representing some typical distortions, the system can deal with the distortions to a large extent.

In this paper, a novel method for SAR image classification based on stationary wavelet transform will be described. First, a SAR image is decomposed into 4 subbands using stationary wavelet transform. Each pixel is then represented by a 4-dimension vector whose components are taken from the wavelet subbands. The pixels are finally classified into a small set of categories by using a parametric supervised classification algorithm. The classification using this wavelet transform was successfully applied to a JERS-1/SAR image.

In this paper, a multispectral image edge detection algorithm is proposed based on the idea that uses global multispectral information to guide local gradient computation. The image is first segmented into a small number of clusters through a clustering algorithm. According to these clusters, a set of linear projection vectors are generated. For a given image, if n clusters are found, there are n(n-1)/2 possible projection vectors. Edge detection is performed by calculating gradient magnitudes separately on each channel. An appropriate projection vector is chosen for each pixel to maximize gradient magnitude. In this way, edges are treated as transitions from one cluster to another. The algorithm has been tested on JERS-1/OPS images, and the experimental results demonstrate its potential usefulness.

Image processing techniques have been used extensively in many different applications today. In particular, in fluid mechanics, image processing has become a powerful technique to study the flow phenomena, the flow pattern and the flow characteristics of two-phase flow. This paper presents a new application of image processing techniques to two-phase bubble/slug flow in a vertical pipe. Based on image processing techniques (image filtering for noise reduction, edge detection and thresholding for image enhancement, etc.), the results obtained are showed that this technique has many advantages. It is able to study together, in very short time, one image contains a large number of bubbles and the large amount of images, while the other methods such as point by point measurements technique or by using a digitizing table for digitization cannot be applicable. Moreover this technique also enable to identify automatically, to measure fast and relatively accurate the parameters such as size, shape of the bubble. These studies promise a great progress for an application of image processing techniques to study the complicated flow phenomena, the flow pattern and the flow characteristics of multiphase flow.

Lossless or high fidelity compression of images is a critical problem yet to be solved in a number of areas such as satellite remote sensing, medical imaging and color image printing. Now the requirement for preservation of image details has rendered the compression method that preserves important information inapplicable. Limited by the storage capacity and transmitting capability, it is very important to enhance the compression ratio of satellite remotely sensed images at high fidelity. Based on wavelet transform and image reconstruction, a feature coding based image compressing algorithm is studied and proposed. This algorithm makes use of the correlativity between the positions of extrema of wavelet transform coefficients as well as the higher-order correlativity between amplitudes of the extrema to perform compression coding, decoding and reconstruction, achieving the result of a compression ratio greater than or equal to 4 at PSNR greater than or equal to 40 db.

In this paper, principles for conversions in pre- and post- processing for HDTV are analyzed, the theoretically ideal processing of the sampling format conversion between YUV4:4:4 and YUV4:2:2, YUV4:2:2 and YUV4:2:0, the size conversion between digital TV standards, and frame rate conversion are presented. The proposed principle is applied in post-processor of HDTV prototype, finally the precision of these post- processing algorithms are reported.

In this paper the line-interlaced scanning DTV broadcasting system is studied. The baseband of interlaced scanning system is introduced which can be reconstructed without spectral aliasing in DTV receiver, and the temporal resolution factor is invited that adjusts the shape of baseband to best match the spectrum of broadcasting video signal. A novel DTV broadcasting system is proposed which employs progressive scanning in camera and in monitor while adopts line-interlaced scanning for video codec, compatible with existing interlaced scanning DTV broadcasting System. This DTV broadcasting system, which only carry the baseband signals of interlaced scanning, can eliminate the spectral replica generated by interlaced scanning, and can reconstruct progressive scanning video signal without spectral aliasing in receiver, hence radically improving the image quality of DTV broadcasting system. Similarly the 2-D interlaced scanning is studied and the improved 2-D interlaced scanning DTV system is proposed. This system is theoretically proved to be able to give visually better image quality of reconstructed picture than the line-interlaced scanning DTV system for it owns larger spectral capacity without aliasing and the optimal 3-D spectrum matching transmission of video sequence.

EvIdent^TM (EVent IDENTification) is a user-friendly, algorithm-rich, graphical environment for detecting, investigating, and visualizing novelty in a set of images. Novelty is identified for a region of interest and its associated characteristics. For functional magnetic resonance imaging, for instance, a characteristic of the region of interest is a time course, which represents the intensity value of voxels over several discrete instances in time. Originally developed for a platform-specific environment using proprietary technology, a new incarnation of EvIdent has been designed using an application programming interface called VIStA^TM (VISualization Through Analysis). VIStA is written in Java^TM and offers a sophisticated generalized data model, an extensible algorithm framework, and a suite of graphical user interface constructs. This paper describes EvIdent and some of its features, the rationale behind the design of VIStA, and the motivations and challenges of scientific programming using Java.

Suppose the image is a realization of a non-stationary random field composed of multiple Gauss-Markov random fields, then the co-occurrence matrix will be able to reflect certain statistical properties of the said random image. Based on this model, the problem of stability of the algorithm for co- occurrence matrix-based image segmentation is raised, certain factors that affect the stability of the performance of segmentation are discussed and methods for enhancing the stability of algorithm are given. As image segmentation is a typical ill-posed problem, there is in general no segmentation criterion and segmentation algorithm that can ensure unique and optimum image segmenting results. For this reason, an autonomous and intelligent segmentation algorithm based on the multi-agents structure is proposed. The correctness and value of application of this method have been proved by experimental results.

This work mainly deal with the processing of Remote Sensed Images gathered from a radiometer flown aboard an environmental satellite. In this work some NOAA/AVHRR images have been used with focus on the southern Europe and the Mediterranean Basin. The images are navigated by resolving the satellite motion and the senors's scanning laws. And calibrated in order to obtain true radiance or temperature values. The first and second band of the AVHRR Radiometer are useful to distinguish between land and sea pixels. While the two thermal infrared channels (4 and 5) proved to be useful to find out the cloud affected pixels and to estimate the value of Sea Surface Temperature on the cloud free pixels. The presented method has been tested on an Intel Pentium 200 MHz based computer. It took 91 sec to perform the following steps: (1) Classify each pixel in one of the following categories: land, sea, cloud, cloud edge, coast line, outside the swath. (2) Estimate on sea, cloudy-free, pixels the Sea Surface Temperature by the means of a low noise split window algorithm on a multispectral image. This performance allows this estimation to be done in real time.

An important process in remote sensing is spectral unmixing which is used to obtain a set of species concentration maps known as abundance images. Linear pixel unmixing, also known as linear mixture modeling, assumes that the spectral signature of each pixel vector is the linear combination of a limited set of fundamental spectral components known as end- members. Thus end-member selection is the crucial first step in the spectral unmixing process. A conveniently parameterized method for determining the appropriate set of end-members for a given set of multispectral images is proposed. The end- members are obtained from a thematic map generated from a modified ISODATA clustering procedure that uses the spectral angle criterion, instead of the common Euclidean distance criterion. The centroids of the compact and well-populated clusters are selected as candidate end-members. The advantages of this technique over common mathematical and manual end- member selection techniques are, (1) the resulting end-members correspond to physically identifiable, and likely pure, species on the ground, (2) the residual error is relatively small, and (3) minimal human interaction time is required. The proposed spectral unmixing procedure was implemented in C and has been successfully applied to test imagery from various platforms including LANDSAT 5 MSS (79 m GSD) and NOAA's AVHRR (1.1 km GSD).

The Lockheed Martin (LM) team had garnered over a decade of operational experience on the U.S. Government's IDEX II (Imagery Dissemination and Exploitation) system. Recently, it set out to create a new commercial product to serve the needs of large-scale imagery archiving and analysis markets worldwide. LM decided to provide a turnkey commercial solution to receive, store, retrieve, process, analyze and disseminate in 'push' or 'pull' modes imagery, data and data products using a variety of sources and formats. LM selected 'best of breed' hardware and software components and adapted and developed its own algorithms to provide added functionality not commercially available elsewhere. The resultant product, Intelligent Library System (ILS)^TM, satisfies requirements for (1) a potentially unbounded, data archive (5000 TB range) (2) automated workflow management for increased user productivity; (3) automatic tracking and management of files stored on shelves; (4) ability to ingest, process and disseminate data volumes with bandwidths ranging up to multi- gigabit per second; (5) access through a thin client-to-server network environment; (6) multiple interactive users needing retrieval of files in seconds from both archived images or in real time, and (7) scalability that maintains information throughput performance as the size of the digital library grows.

On board Image compression is a very powerful tool to optimize the onboard resources needed to store and transmit image data to ground. This is due to the steady performance improvements of the compression algorithms and to the availability, for spaceborne applications, of highly integrated circuits (ASIC technology) that made it possible to implement very sophisticated real-time schemes. We propose in this paper a survey of on on-board image compression with emphasis on some compression architectures and present the future prospects.

The solar vector magnetograph system in Huai Rou Solar Observation station (HRSO) is described in this paper. First the measuring principle of magnetic field and velocity field are discussed. We take the simple two-passband birefringent filter as an example to explain the working principle of the narrow-band birefringent filter which is the core of acquiring magnetic field and velocity field. Then the 35 cm Solar Magnetic Field Telescope is described, including optic system, camera system and digital system. In camera system the advantages of scientific Charge Coupled Devices (CCD) are described. In digital system we introduce the necessary functions of image grabber which is suitable for acquiring solar magnetic field. At last the performance of the whole system is discussed such as resolution and signal-to-noise ratio, and one magnetic field image and one velocity field image in the same area on the Sun are shown.

This paper describes a prototype telemedicine system for remote 3D radiation treatment planning. Due to voluminous medical image data and image streams generated in interactive frame rate involved in the application, the importance of deploying adjustable lossy to lossless compression techniques is emphasized in order to achieve acceptable performance via various kinds of communication networks. In particular, the compression of the data substantially reduces the transmission time and therefore allows large-scale radiation distribution simulation and interactive volume visualization using remote supercomputing resources in a timely fashion. The compression algorithms currently used in the software we developed are JPEG and H.263 lossy methods and Lempel-Ziv (LZ77) lossless methods. Both objective and subjective assessment of the effect of lossy compression methods on the volume data are conducted. Favorable results are obtained showing that substantial compression ratio is achievable within distortion tolerance. From our experience, we conclude that 30dB (PSNR) is about the lower bound to achieve acceptable quality when applying lossy compression to anatomy volume data (e.g. CT). For computer simulated data, much higher PSNR (up to 100dB) is expectable. This work not only introduces such novel approach for delivering medical services that will have significant impact on the existing cooperative image-based services, but also provides a platform for the physicians to assess the effects of lossy compression techniques on the diagnostic and aesthetic appearance of medical imaging.

Remote sensing images acquired in various spectral bands are used to estimate certain geophysical parameters or detect the presence or extent of geophysical phenomena. In general, the raw image acquired by the sensor is processed using various operations such as filtering, compression, enhancement, etc. in order to enhance the utility of the image for a particular application. In performing these operations, the analyst is attempting to maximize the information content in the image to fulfill the end objective. The information content in a remotely sensed image for a specific application is greatly dependent on the gray-scale resolution of the image. Intuitively, as the gray-scale resolution is degraded, the information content of the image is expected to reduce. However, the exact relationship between these parameters is not very clear. For example, while the digital number (DN) of a pixel may change as a result of the decrease in the number of gray scales, it may be possible that the overall image classification accuracy (a measure of information content) may not show a corresponding reduction. Furthermore, the degradation in information content has been shown to be related to the spatial resolution also. Our simulation studies reveal that the information content does indeed drop as the gray-scale resolution degrades. Similar results are observed on working with real images. We have developed a simple mathematical model relating the image information content to the gray-scale resolution, using which the optimal number of gray scales necessary to interpret an image for a particular application may be deduced.

This paper presents a contrasted statistical processing approach to obtain improved probabilities of false alarm when performing automatic target detection. The technique is based upon analyzing each sector of the image and comparing it with surrounding windows in which the desired statistical property is calculated. The contrast of the statistical property is extracted using the prediction or the prediction-correction equations. The contrast of the statistical property is shown to be a good discriminator of the target from its background allowing the reduction of the detection threshold applied over the stationary region while maintaining a constant false alarm probability.