A special NIST database of live-scan fingerprint with elastic distortion wa prepared. It is used to evaluate the effect of elastic and other distortions on correlation filters. The need for normalized and fine vs. coarse rotationally-aligned data are addressed with performance gains for various cases noted. Procedures to test and evaluate fingerprint recognition algorithms for verification and identification are defined for the first time and initial results are presented.

Biometric verification refers to the process of matching an input biometric to stored biometric information. In particular, biometric verification refers to matching the live biometric input from an individual to the stored biometric template of that individual. Examples of biometrics include face images, fingerprint images, iris images, retinal scans, etc. Thus, image processing techniques prove useful in biometric recognition. In particular, composite correlation filters have proven to be effective. In this paper, we will discuss the application of composite correlation filters to biometric verification.

We propose to use integral images to reconstruct and recognize three-dimensional (3D) scenes in a computer. A stereo-matching algorithm is applied to integral images in order to extract the depth information. This information is used to digitally reconstruct the 3D scenes. A numerical 3D correlation is then computed between various reconstructed scenes. We demonstrate the reconstruction and correlation results from experimental integral images. We propose to use a nonlinear correlation for better discrimination and we present the successful recognition and 3D localization of an object in a 3D scene. We finally compare the discrimination of two- and three-dimensional correlations.

Advanced Optical Systems has developed the world's smallest and lowest cost, fully functional target recognition and tracking system. The heart of the ULTOR target recognition and tracking system is an optical correlator. The system includes real-time preprocessing, large filter stores, filter management logic, correlation detection and thresholding, correlation tracking, and data output. It is self contained, receiving operational commands as an Internet appliance. We will present a demonstration of some of the capabilities of the system using live video signals and real target models. The ULTOR system has wide application in both military and commercial settings. The Navy is considering use of the ULTOR system in several programs, including missile systems and unmanned aerial vehicles.

JPL has recently developed, for the first time, a compact (2” x 2” x 1”) Grayscale Optical Correlator (GOC) using a pair of 512 x 512 Ferroelectric Liquid Crystal Spatial Light Modulators. In this paper, we will discuss recent progress in the design and packaging technology to achieve a rugged portable GOC module to enable the real-time onboard applications of this miniature GOC. Several automatic target recognition applications will also be presented.

We propose the use of a vision-guided system for the formation and dynamic manipulation of an assembly of microscopic particles. Microscopic particles having visually recognizable characteristics, i.e. shape and color, are arranged and manipulated using patented and recently demonstrated fully dynamic multiple-beam optical trapping for real-time and simultaneous manipulation and control of arrays of particles. A derived application is the simultaneous manipulation of an assembly of particles at the same time as they are observed and identified with microscopic image processing. We will point to novel experiments based on micro-fluidic and lab-on-a-chip technology, where the aim is to attain all-optical and simultaneous control of aggregated micro-structures that can function as channels, reservoirs, sensors and micro-pumps.

Linear correlation filters have been implemented in optical correlators and successfully used for a variety of applications. The output of an optical correlator is usually sensed using a square law device (such as a CCD array) which forces the output to be the squared magnitude of the desired correlation. It is however not a traditional practice to factor the effect of the square-law detector in the design of the linear correlation filters. In fact, the input-output relationship of an optical correlator is more accurately modeled as a quadratic operation than a linear operation. Quadratic correlation filters (QCFs) operate directly on the image data without the need for feature extraction or segmentation. In this sense, the QCFs retain the main advantages of conventional linear correlation filters while offering significant improvements in other respects. Not only is more processing required to detect peaks in the outputs of multiple linear filters, but choosing a winner among them is an error prone task. In contrast, all channels in a QCF work together to optimize the same performance metric and produce a combined output that leads to considerable simplification of the post-processing. In this paper, we propose a novel approach to the design of quadratic correlation based on the Fukunaga Koontz transform. Although quadratic filters are known to be optimum when the data is Gaussian, it is expected that they will perform as well as or better than linear filters in general. Preliminary performance results are provided that show that quadratic correlation filters perform better than their linear counterparts.

Composite correlation filters have been demonstrated in many synthetic aperture radar (SAR) automatic target recognition (ATR) applications because of their ability for class discrimination and distortion-tolerance with shift invariance. For many implementations, a simple model of white noise spectral density has been used to reduce output noise variance. Substituting a colored noise model or real clutter imagery in synthesizing correlation filters has shown improved clutter rejection and target recognition. However, the spectral response of SAR imagery is quite different than other types of images. We demonstrate the use of SAR clutter imagery and a new model for SAR clutter estimation for use in maximum average correlation height (MACH) and optimal tradeoff distance classifier correlation filter (OTDCCF) approaches. Test results using the MSTAR database are shown.

In previous work we introduced new algorithms for forcing two dissimilar patterns to correlate with each other with very high discrimination capability. In this work we analyze our algorithms mathematically using nonlinear-transform methods. In this analysis we distinguish what terms are responsible for the correlation and what terms are responsible for the heterocorrelation. Computer simulations supporting our mathematical analysis are presented.

The various types of synthetic discriminant function (sdf) filter result in a weighted linear superposition of the training set images. Neural network training procedures result in a non-linear superposition of the training set images or, effectively, a feature extraction process, which leads to better interpolation properties than achievable with the sdf filter. However, generally, shift invariance is lost since a data dependant non-linear weighting function is incorporated in the input data window. As a compromise, we train a non-linear superposition filter via neural network methods with the constraint of a linear input to allow for shift invariance. The filter can then be used in a frequency domain based optical correlator. Simulation results are presented that demonstrate the improved training set interpolation achieved by the non-linear filter as compared to a linear superposition filter.

The projection-slice synthetic discriminant function filter (PSDF) synthesis is discussed with consideration of individual projection energy in order to improve the performance of the filter. This improvement is made relative to the normal PSDF filter and the basic synthetic discriminant function filter. The results of application of these new techniques to the processing of real data is reported and evaluated using accepted metrics such as the Fisher discrimination metric.

One of the main problems of optical data processing is the problem of image recognition. There were given much attention to optoelectronic methods of recognition of distorted images nowadays. There are a number of different approaches for the solution of such problem. One of the most popular approaches is using of optical correlators for this field. The main problem of this approach is to select an object to provide a correlation of input image with it. One of the widespread methods is to use an effective object-an invariant correlation filter. The paper presents the results of investigations on image recognition with the help of Wavelet Basis Kernel Filters (WBKF). Both results of the theory and computer simulations are presented. Also computer simulations hold a comparison of image recognition results with the help of other different approaches (GMACE, SDF and so on). The obtained results seem to be better for WBKF recognition in some cases. There are presented authors suggestions about using of WBKF filters for different distortion invariant image recognition problems and results of image recognition in presence of white noise.

We consider new distortion-invariant filters (DIFs) to detect objects in high-resolution Electro-Optical (EO) visible imagery. EO data is a difficult detection problem, because only primitive features such as edges and corners are useful. No hot spots (present in IR data) or bright reflectors (present in SAR data) exist in EO data. We thus expect many false alarms when we try to detect objects in EO data. We use new eigen-detection filters because they are shift-invariant, require only few filters and can handle multiple target classes. Initial results show that our filters, when using zero-mean data, perform well on EO data.

Improvements in silicon foundry processes have made possible high-resolution, light-efficient backplanes capable of driving electro-optic modulators with higher voltage signals. The higher voltage provides the excitation to achieve sub-millisecond response times with a wave of phase modulation when used with dual-frequency nematic liquid crystals. By combining dual-frequency phase modulators with high-voltage silicon backplanes, compact spatial light modulators become available for applications that need fast, high-throughput modulators such as optical signal processing, adaptive wavefront correction, optical signal routing or beamsteering, and active diffractive optics.

The precision of a radial basis function (RBF) neural network based tracking method has been assessed against real targets. Intensity profile feature extraction was used to build a model in real time, evolving with the target. Precision was assessed against traditionally measured frame-by-frame measurements from the recorded data set. The results show the potential limit for the technique and reveal intricacies associated with empirical data not necessarily observed in simulations.

A Hopfield net is a feed-back neural network (FBNN) consisting of one layer of binary neurons. Its main function is its ability to associatively store multiple binary patterns (or accurate binary patterns) in its connection matrix, and to associatively recall on any of these stored patterns by a nosie affected triggering pattern applied to the input. If the noise of this input pattern falls within a certain range, (called the domain of convergence or the domain of attraction in the terminology of nonlinear system analysis), of a certain accurately stored pattern, then that accurate pattern will be recalled and will appear permanently in the output of the FBNN even when the triggering input pattern is removed. This being so is due to the self-sustained feedback action and the domain of convergence properties existing in the FBNN. However, if the stored patterns and the noisy, triggering patterns violate some PLI (positive, linear, independency) condition, then it is impossible for the FBNN to learn the mapping relations of these accurate patterns to be triggered by the triggering patterns with the designated noisy range. In this case we have to design a "UNIVERSAL" two-layered feed-back neural network that will accomplish this learning task. This paper dervies from the principle of the NONITERATIVE LEARNING, the design of a universal, parallel-cascaded two-layered perceptron (PCTLP), and the reconnection of it to form a universal FBNN, (which may be called the generalized Hopfield net,) that will accomplish this associative-storage and associative-recall learning task.

An approach to target tracking is presented that utilizes adaptive gray-scale correlation. The algorithm is implemented in software and executed in real-time on commercial-off-the-shelf image processing hardware. The basic correlation scheme utilizes non-adaptive commercial library calls that rely on image pyramids for speed. A framework is described for implementing a general-purpose adaptive correlation capability. The approach is robust and tolerant to scenarios involving rotation, scale changes, and contrast reversal. The algorithm also solves problems associated with walk-off.

White Sands Missile Range has developed a video tracker using a consumer grade personal computer and frame grabber. The video tracker receives NTSC video at a 60Hz field rate and delivers tracking data with a 32ms delay. Thirty video trackers have been fielded with great success tracking highly dynamic rockets and airplanes from launch with an azimuth slew rate of up to 60 degrees/second. Simple contrast tracking is employed for both initial launch and steady-state tracking. Smart region of interest determination is used to facilitate servo control issues.

Future Mars/planets explorations call for precision and even pinpoint landing. Low cost optical correlator is one of the promising enabling technologies for pinpoint landing. JPL has developed a state-of-the-art miniature optical correlator (MOC) to demonstrate its feasibility. In this paper, we describe a simulation testbed under development for measuring MOC’s performance in a high-fidelity entry, descent, and landing environment, and provide our preliminary simulation result.

We propose and demonstrate a photorefractive real-time holographic deconvolution technique for adaptive one-way image transmission through aberrating media. In contrast with preceding methods, which have typically required various coding of the exact phase or two way image transmission for correcting phase distortion, our technique relies on one-way image transmission through using exact phase information. Our technique can simultaneously correct both amplitude and phase distortions and provide substantial noise filtering. The nonlinearity of the photorefractive medium also helps to enhance the signal-to-noise ratio (SNR). And is thus superior to previous methods. We demonstrate our results through both experiment and computer simulation for different aberrators.

It often happens that the number of samples available to train a discriminator is many fewer than Learning Theory tells us we need to accomplish the required accuracy/confidence. When you run up against a theoretical limit, only two choices are possible. You can accept the situation, or you can look for ways around those limits. This report suggests that there is a way around conventional learning theory and applies the new technique (called Margin Setting) to a difficult artificial problem to illustrate its power.

Optical Mark Recognition (OMR) is thoroughly reliable and highly efficient provided that high standards are maintained at both the planning and implementation stages. It is necessary to ensure that OMR forms are designed with due attention to data integrity checks, the best use is made of features built into the OMR, used data integrity is checked before the data is processed and data is validated before it is processed. This paper describes the design and implementation of an OMR prototype system for marking multiple-choice tests automatically. Parameter testing is carried out before the platform and the multiple-choice answer sheet has been designed. Position recognition and position verification methods have been developed and implemented in an intelligent line scan camera. The position recognition process is implemented into a Field Programmable Gate Array (FPGA), whereas the verification process is implemented into a micro-controller. The verified results are then sent to the Graphical User Interface (GUI) for answers checking and statistical analysis. At the end of the paper the proposed OMR system will be compared with commercially available system on the market.

We show multiple uses of space-time Fourier analysis in segmenting particular "activities" from sequences of images. Examples of such analysis include the detection and characterization or mitigation of scintillations, harmonic motion, and transverse motions of objects. Characterizations can include the estimation of parameters such as oscillation frequencies and distances and velocities of moving objects. We make use of stereo and time sequence imaging to generate scene data of higher spatio-temporal dimension than two. We demonstrate purely digital as well as hybrid digital/optical (correlator) implementations and discuss techniques for mappings between space and time in utilizing imaging and optical resources.

A Phase-only spatial light modulator can provide active spot pattern projection with high signal-to-noise ratio and form near-arbitrary phase modulation surfaces. As a result they can diffract laser beams into a near-arbitrary pattern of laser spots. Depending on the sequence of phase images loaded onto the SLM, the spots can be scanned on independent and continuous two-dimensional trajectories. We refer to this flexible beamsteering system as the real-time adaptive multi-spot laser beamsteering system (RAMS-LBS). This paper presents work under progress, in developing 2D and 3D calibration algorithms for a spot pattern projection system. In the 2D calibration process, spot grids are projected with successively more spot locations. After each projection, a higher order model is determined for camera to projector coordinate transforms. The accuracies of different model orders are measured. In the 3D calibration process, grids of spots are projected onto non-coplanar target grid to construct the transformation matrix between different coordinates. Perspective distortions are included in the transformation vectors after the calibration. Therefore, 3D information of the target can be obtained in the calibrated system. Applications such as 3D target surface topology measurement and target detection using 2D and 3D information are described in this paper.

In our previous paper "Theoretical analysis and simulation of heterocorrelation filter" we attempted to find the best conditions that makes the relative intensity of the autocorrelation to be equal to the heterocorrelation. Our theoretical modeling and computer simulation based on the theory developed in the previous papers showed that in order to make the autocorrelation equal to the heterocorrelation, it is essential to have phase-only based heterocorrelation.

Hyperspectral (HS) data contains spectral response information that provides detailed chemical, moisture, and other descriptions of constituent parts of an item. These new sensor data are useful in USDA product inspection and in automatic target recognition (ATR) applications. However, such data introduces problems such as the curse of dimensionality, the need to reduce the number of features used to accommodate realistic small training set sizes, and the need to employ discriminatory features and still achieve good generalization (comparable training and test set performance). HS produces high-dimensional data; this is characterized by a training set size (N_i) per class that is less than the number of input features (HS λ bands). A new high-dimensional generalized discriminant (HDGD) feature extraction algorithm and a new high-dimensional branch and bound (HDBB) feature selection algorithm are described and compared to other feature reduction methods for two HS product inspection applications. Cross-validation methods, not using the test set, select algorithm parameters.

A study was conducted on the use of morphological processing for detection in Uncooled Infra Red (UCIR) data using the Clutter type 1 and Clutter type 2 databases. The CMO algorithm was used with various modifications. A fixed minimum peak value was used rather than a fraction of the maximum peak per image. This is much more realistic, since many real scenes will not contain targets. One-dimensional directional structuring elements (SEs) were used for the Close Minus Open algorithm. We used the range of gray levels within the output blob peaks (blob analysis) and larger windows in peak sorting to reduce false alarms. A new dilation minus erosion morphological algorithm gave the best result.

The new methods of invariant pattern recognition (IPR), based on the effective calculation of image moment features are presented. It is described the special purpose multiprocessor computer system, which realizes the proposed methods of IPR. The system is reconfigurable, with the architecture controlled by parameters of the input images.

A novel polarization-based fingerprint identification sensor is proposed in this paper. This sensor consists of an optoelectronic system where the enrollment process is recorded optically and the identification process is carried out digitally using the concept of fringe-adjusted joint transform correlation technique. In the optical part, a polarized coherent light beam is used to record spatially dependent response of the scattering medium of the fingerprint to provide detailed surface information, which is not accessible to mere intensity measurement. Both simulation and experimental results are presented to evaluate the performance of the proposed technique.

A new feature extraction algorithm is proposed for content based image retrieval. The new features can capture both local and global information at high speed. Thus, it is suitable for large-scale image retrieval. We demonstrate both texture image retrieval and real-world scene retrieval. Our results show that the cepstral based texture features provide an efficient and effective tool for content-based image retrieval.

Computer generated hologram making by laser phototypesetting device is considered. Laser phototypesetting device is chosen to print computer generated holograms due to its great resolution (up to 5000 dpi) in comparison with laser printer. Experimental values of film phase distortions for computer generated hologram record are given. Computer generated hologram synthesis results and experimental results of such holograms making by laser phototypesetting device are discussed. Characteristics of this class holographic filters and possibilities of their use in diffraction correlators with spatially coherent and incoherent light are estimated.

There are presented the results of the investigations of the fingerprints' images correlation recognition in conditions of different distortions -- scale, angular orientation change, image's surface reducing, noises' influence. There are examined possibilities of the person's identification and their verification. There are proposed and investigated the method of the fingerprints' semi-spectrums recognition and the method of the fingerprints' space-dependent recognition.

There are presented the structures of the special purpose mono-channel and multi-channel optical-electronic systems and are described computing processes in the systems at the realization of the different fingerprints recognition algorithms: "FSR-1," "FSR-2," "FSDR-1," "FSDR-2," "FICR." Also, there are presented the results of systems investigations: fingerprints time recognition, systems productivity at the fingerprints comparison step, systems prices.

The new image complexity informative feature is proposed. The experimental estimation of the image complexity is carried out. There are elaborated two optical-electronic processors for image complexity calculation. The determination of the necessary number of the image's digitization elements depending the image complexity was carried out. The accuracy of the image complexity feature calculation was made.

As a new approach to implement a function of the conventional lenticular sheet used in the autostereoscopic 3D display system, a VHLS (volume holographic lenticular sheet) is suggested. This VHLS is made by recording the MxN (M: number of view, N: number of window pitch) vertical stripe patterns on a volume holographic recording material using an angular multiplexing method. In the experiment, 4-view VHLS is manufactured by using a Dupont photopolymer (HRF-150-100) and then, a VHLS-based autostereoscopic 3D display system is implemented. 4-view strip patterns of 4 x 256 pixels are synthesized to be an image of 1024 x 768 pixels in GUI (graphic user interface) environment, in which the synthesized image consists of 256 pitches and each pitch is composed of 4 pixels. Then, each strip pattern is sequentially recorded in the photopolymer with the corresponding angular-multiplexed reference beams. Some experimenal results show that 4-view imge patterns are diffracted from a synthesized image through the VHLS and these beams are spatially separated each other on the screen, so that a possibility of implementing a new VHLS-based multiview autostereoscopic 3D display system is suggested.