Electronic mail is a facility, analogous to postal mail, in which computers are used to compose, deliver, and receive messages. Traditional electronic mail systems rely solely on text as the medium of communication. A multi-media electronic mall application, Mail, combines the media of Rich Text, voice, images, and electronic documents to facilitate interpersonal communication. With Mail, these various media can be integrated into a single message. The variety of available media and the complexity of the messages that result from their combination make it important for Mail to have a simple user interface. It was possible to develop a simple, graphically-based interface that would accommodate Mail's message complexity. The integration of these diverse media is made practical by the rich operating environment in which Mail runs. Modern advances in hardware, operating systems, libraries, and servers make possible this powerful multi-media electronic mail application.

Convolution mask is a basic and useful tool for image processing allowing edge detection, noise cleaning, texture extraction . . . Nevertheless, such operators are time consuming in the two following cases : when the mask is large (until 64*64 for texture or correlation extraction ) and when the convolution is iterated. A parallel architecture is efficient in order to limit the processing time but necessitates an overlapping (2n lines and 2n columns) of the fields affected to the different processors. Such an overlapping allows a limited number of iterations ( n in the case of a 3*3 mask) . Thus, if we need more than n iterations, a refreshment of the memory is done in order to process only " true " radiometries.

There are a variety of emerging technologies which will facilitate the incorporation of video into the work station environment. These technologies, whether analog or digital, very often require the addition of special purpose hardware to be added to the work station or personal computer. We have been investigating algorithms for the display and integration of video image sequences into a PC using an off-the-shelf graphics adapter. The techniques we employ are not as powerful as those taking advantage of special purpose hardware. In particular, no provisions are made for either real-time capture or full screen motion. However, by providing the simultaneous display of multiple, scalable, moving movies, "multi-media" applications which use preprocessed, window-sized video sequences can be designed. In this paper we describe both techniques for image reduction and coding and various strategies for decoding, playing, and smoothly moving video image sequences.

Bit-mapped drawings are normally created by using the so-called "Paint Editor" through the provided image primitives. A set of operation tools are also available for interactive manipulation of primitive objects. Window-based operating environments, such as the IBM OS/2 Presentation Manager,1M provide an Application Programming Interface to send messages to modules that actually carry out bitmap generation. We observed that this message-driven architecture can facilitate the "capture and reapplication of user operations" for application programs. Therefore, it is possible to create a set of bit-mapped drawings that conform to certain properties through the same sequence of operations. Two applications are described in this paper. The first is the creation of fonts with a consistent typeface. The sequence of operations used in designing the typeface is automatically recorded. The user can then generate bitmaps of other characters with the same typeface by repeating the same sequence of operations. The second application is the generation of shadows of 3D objects. If a profile of an object is given, the shadow can be synthesized by bitmap manipulation primitives. When the same sequence of primitives is applied to different objects, shadows can be generated correctly.

An approach to integration of software and hardware for image production closely coupled with real-time control of an instrument is described. Image production includes reconstruction, processing, display, archiving and export. Feedback of information from images to instrument control is an essential feature, often with a human operator in the loop. Real-time control requires guaranteed response to external events, support for multiple processors and a variety of interfaces to instrumentation. Image production involves implementation of reconstruction and processing algorithms, a windows system, and network support. Both functions need a consistent user interface and an adaptable programming environment . The UNIX operating system running on a network of workstations meets the needs of image production and is very well suited to user interface and software development for both imaging and control. However, high performance real-time control is not possible under most available UNIX systems, but can be accomplished using a real-time kernel running on separate hardware. This suggests a multi-processor approach based on inter-process communication between the workstation and control environments. The architecture consists of three hardware layers (UNIX workstations, board-level microcomputers, and digital signal processors) and four software layers. The real-time software layers are DSP microcode and the microprocessor code. The lower layer of workstation software is a multi-tasked command-driven program tailored for instrument control and image production. The top layer of workstation software consists of an icon or menu driven user interface. A research instrument for nuclear magnetic resonance imaging will serve as an example of this approach.

The ideal image processing system would be one that people already know how to use. Its components would be interchangeable with other computers in the office and in the factory. It could be quickly programmed to do its job by nonprogrammers and non-typists. And it would communicate and print reports over the same network used by all the other computers in your company. Does this sound like science fiction? It's not. The image processing system in this scenario represents a trend toward standardization in the industry. Image processing suppliers are realizing that they must move away from their existing proprietary architectures and customized software systems in favor of standard computers having a wide base of support hardware and software, and a wider pool of knowledgeable users.

We consider what is needed to create electronic document libraries which mimic physical collections of books, papers, and other media. The quantitative measures of merit for personal workstations-cost, speed, size of volatile and persistent storage-will improve by at least an order ofmagnitude in the next decade. Every professional worker will be able to afford a very powerful machine, but databases and libraries are not really economical and useful unless they are shared. We therefore see a two-tier world emerging, in which custodians of information make it available to network-attached workstations. A client-server model is the natural description of this world. In collaboration with several state governments, we have considered what would be needed to replace paper-based record management for a dozen different applications. We find that a professional worker can anticipate most data needs and that (s)he is interested in each clump of data for a period of days to months. We further find that only a small fraction of any collection will be used in any period. Given expected bandwidths, data sizes, search times and costs, and other such parameters, an effective strategy to support user interaction is to bring large clumps from their sources, to transform them into convenient representations, and only then start whatever investigation is intended. A system-managed hierarchy of caches and archives is indicated. Each library is a combination of a catalog and a collection, and each stored item has a primary instance which is the standard by which the correctness of any copy is judged. Catalog records mostly refer to 1 to 3 stored items. Weighted by the number of bytes to be stored, immutable data dominate collections. These characteristics affect how consistency, currency, and access control of replicas distributed in the network should be managed. We present the large features of a design for network docun1ent/image library services. A prototype is being built for State of California pilot applications. The design allows library servers in any environment with an ANSI SQL database; clients execute in any environment; conimunications are with either TCP/IP or SNA LU 6.2.

An image based geographic information system (GIS) using an efficient combination of image (raster) data and vector data has been developed. The system has been realized on a 32 bit engineering work station (TOSHIBA AS3000 series) with special hardware for high speed image data retrieval and display. The system adopts several (up to four) 5.25 inch optical disks to store the image data, because of compactness and removability. It uses image data for a base map in combination with vector data and attribute data (text or numeral data), while the conventional GIS uses vector data for the base map because of its lack of image processing capability. Two new techniques employed in the system are described. One is a technique to obtain a seamless base map from a conventional printed atlas. It is necessary to adjust the mismatching caused by the distortion between two neighboring maps. The other is a technique for distributingC image data to multi-optical-disks and concurrently retrieving from them. It has proved that the proposed image handling techniques reduce the initial base map input cost, and make the man-machine interface powerful, with such functions as a smooth scmll, movies, color photographs, and a variety' of visual information.

Document scanning is now an accepted part of office procedure, allowing the incorporation of digitized images into new documents and the conversion of scanned print into ASCII by optical character recognition ( OCR). Often document pages contain more than one form of information - textual, graphical and/or pictorial. Segmentation of document images into these three categories is feasible with the aid of image processing. Projections of the thresholded document images in conjunction with autocorrelation are used to check text alignment. Then the edge shifting properties of the rank filter are used to coalesce image regions containing text into solid near-rectangular blocks. Pyramidal reduction is combined with the filtering to ease the computational burden. Horizontal and vertical projections are used to segment whole pages recursively into homogeneous blocks whose properties are then analysed. Applications forseen for the image segmentation include modified facsimile systems, achievement of artifact-free OCR and conversion of document images into files with separate formats for text, graphics and pictures.

A paradigm for the visualization of image interpretation systems is described. The paradigm uses visual interactions with the human observer to extract visual knowledge from the user and format it into rules, thresholds and choice of relevant attributes. The detailed description of the algorithm is presented through the analysis of an example of grouping line-segments into straight lines. Two different levels of abstraction are involved in the proposed learning mechanism. One involves the calculation of values and thresholds and is based on a statistical analysis of the user's chosen examples. The other mechanism deals with reformatting and rewriting of rules and is a symbolic process that belongs to the more abstract levels of interpretation. The mechanism for reconsideration incorporates the two concept learning paradigms, extension and correction. The correcting of formerly learned Rules is based on additional examples chosen by the user and invokes a new, low level, calculation of thresholds only as a last resource.

This paper introduces a general purpose method for black and white image segmentation, which is based on the design of a rule based-system. Rules integrate general knowledge, which is completely independent of environment and scene content ; they process simultanuously regions and lines by merging similar regions, splitting non uniform ones, connecting lines or deleting non significant ones. Rules are selected according to the nature of data under analysis, evaluated by means of a set of performance measurement and ordered according to their efficiency ; only the most efficient ones are really fired. Furthermore, adapted control strategies, using dynamic data selection, allow to focuse the process on the required parts of the image. This paper describes some examples of the rule, and the set of performance measurement for rule efficiency evaluation.

By analyzing the tracking scenario thoroughly, the gate tracking scenario may fall into four states:(Si) Common tracking state:main state in which the target moves smoothly and there is little disturbance of foreground and background. (52) the target unstable state in the gate:the target violent movement and the target existence in strong disturbances. (83)The taret is escaping from the tracking gate. (S4)the target is completely Iost.Fuzzy mathematics is introduced to describe those four states and transforms among them.The finite state fuzzy automachine is formed. Several statistical parameters have been proved to be very effective to describe the four states by mean of fuzzy concepts. The automachine can provide both the fuzzy probabilities of four tracking states and the transform matrix among them. The elements of the matrix which present the transform probabilities among the four states are set up by the combination of statistical parameters. The fuzzy automachine can provide the most probable state decision to guide the multi-state gate tracker to chose the proper tracking tactics at any time. The different tracking algorithm can be applied to different state in order to keep on tracking under both the common tracking situation and uncommon tracking situation6.

Region analysis of binary images includes standard "blob counting" as well as the computation of geometric properties (area, perimeter, moments, number of holes) of the connected components or regions in an image. Software algorithms for region analysis are well known, and special purpose vendor systems are now available for real-time and near real-time binary region analysis. In this paper, we describe algorithms using a combination of software and inexpensive image hardware (e.g. boards from Matrox, Imaging Technology) that perform region analysis on 512x480 images in times ranging from about a half second ( < 10 regions) to a few seconds (several hundred regions) on a PC/AT (80286) based system, and well under a second for all the cases we have tested on a 80386 based system.

This paper proposes a feedback process approach to shape transformation in OCR character recognition for achieving higher recognition rates. An input image is deformed by warping function so that the stroke in the input image and the corresponding stroke in the reference image will overlap each other as much as possible. Pattern matching between both images is accomplished at several blurring levels, and a gradient method is applied to obtain the optimal warping parameters for minimizing the distance between the images in each blurring level. Third order polynomials by expanding Affine transformations are used as warping functions. Two kinds of pattern matching experiments were carried out. First, the image generated from an original image by warping function with certain parameters was matched with the original image by the proposed method. Second, machine-printed character images of different fonts were used for pattern matching. Both experimental results show that the optimal warping parameters for global minimization of the distances were successfully determined by the combination of a coarse-to-fine pattern matching and a gradient method. The proposed method is applicable to a variety of shape compensations, based on other matching criteria.

Character recognition methods can be categorized into two major approaches. One is pattern matching, which is little affected by topological changes such as breaks in strokes. The other is structural analysis, which tolerates distorted characters only if the topological features of their undistorted versions are kept. We developed a new recognition method for hand-written numerals by combining the merits of the two approaches. The recognition process consists of three steps: (1) an input character is recognized by a patternmatching method, which reduces the number of possible categories to 1.5 on the average, (2) the character is yenfled to be true, false, or uncertain by a structural analysis method that we have newly developed, and (3) special heuristic verification logics are applied to uncertain characters. In the second step, the new structural analysis method uses the positions and directions of terminal points extracted from thinned character images as a main feature. The extracted terminal points are labeled according to a structural-feature distribution map prepared for each category. The generated labels are matched with template label sets constructed by statistical analysis. The characteristics of the method are as follows: (1) it copes with distortion of hand-written characters by using distribution maps for the positions and directions of feature points, and (2) distribution maps can be automatically generated from statistical data in learning samples and easily tuned interactively. The merits of combining the two methods are as follows: (1) the advantages of both pattern matching and structural analysis are obtained, (2) the probabilities of steps 2 and 3 needing to be executed are 22% and 9% respectively, which hardly affect the total processing time, and (3) as a result of steps 1 and 2, only a small number of special logics are required. In a test using unconstrained hand-written characters of low quality, the recognition rate and substitution rate were 95.2% and 0.42% respectively. A recognition speed of 80 characters/second was achieved on a small hardware system.

Computer systems thatprocluce color images, usually consist ofcombinations ofhardware and software components that perform different functions, such as capturing, synthesizing or editing images, incorporating images into documents, proofing, and rendering results. Images, and the documents containing them, must be stored temporarily or archivally, and transmitted from component to component. Users will be able to operate such systems more conveniently and flexibility if the parts communicate with each other using a standard interchange format, allowing any conforming module to communicate with any other. This paper discusses the requirements for such an interchange color space, comparing them to some ofthe criteria used to measure traditional color spaces. Any computer interchange space should employ the principles of colorimetry to provide deviceindependence, which is necessary if negotiation between components is to be avoided. Other requirements are accuracy and computational efficiency of transforms to and from device-dependent and internationally standard spaces, ability to represent all visible colors, maximizing the amount of space occupied by the most likely image color gamuts, and robustness against quantization and roundoff errors. This paper proposes ways to measure the performance of color spaces against the defined requirements and applies the tests to several well-known color spaces.

This paper presents a method to automatically obtain a contour representation of character patterns and rapidly generate high-quality characters of various sizes. Contour representation generates contour vectors and attribute data. The contour vectors are generated without losing information about the characters' shape. The attribute data is used to correct non-uniform widths of vertical and horizontal lines, which is often a problem when characters are transformed to lower resolution. The character generation step reproduces the characters using data obtained in the contour representation step. Since this method represents characters by contours, it does not require as much data. And since non-uniform line widths are corrected rapidly, the characters are of high quality.

In this paper, a technique to convert page description language (PDL) code to a bitmap color image using an optical image memory is discussed. The technique is useful in developing a low cost PDL compatible high resolution graphic printing device for desktop publishing applications. The optical image memory in this discussion is a raster scan addressable device which is write-only in the sense of electrical digital access, and its content is continually accessible by an analog optical mean. Utilizing the low cost optical image memory, the algorithm can rasterize PDL code without large amounts of expensive digital image frame memory. Since the optical image memory is a raster scan addressable device, the rasterization ofeach pixel ofthe color image must be done within the access time of each pixel writing. Some novel techniques are introduced in this algorithm. System development, performance analysis, and the application of the algorithm will be presented.

It has become increasingly important to get high quality prints from TV or other video signals in the desktop publishing and multi-media environment. Normally, a video signal has very low spatial resolution compared with the spatial resolution of printing devices. Hard copy from direct printing of a video signal is usually very poor in quality. To improve the video signal spatial resolution for printing, the algorithm presented in this paper applies an adaptive spatial interpolation technique to the video signal. The algorithm is based on a computer graphics curve-fitting algorithm with the addition of edge-restricted interpolation to sharpen edges and to enhance image details. Therefore, the new interpolation algorithm will not only improve the spatial resolution but also improve the sharpness of the printed image. In this paper, the aspects of theoretical development of the new algorithm will be discussed and some simulation results will be shown and evaluated to present the effect of the algorithm.

A regional information guidance system has been developed on an image workstation. Two main features of this system are hypermedia data structure and friendly visual interface realized by the full-color frame memory system. As the hypermedia data structure manages regional information such as maps, pictures and explanations of points of interest, users can retrieve those information one by one, next to next according to their interest change. For example, users can retrieve explanation of a picture through the link between pictures and text explanations. Users can also traverse from one document to another by using keywords as cross reference indices. The second feature is to utilize a full-color, high resolution and wide space frame memory for visual interface design. This frame memory system enables real-time operation of image data and natural scene representation. The system also provides half tone representing function which enables fade-in/out presentations. This fade-in/out functions used in displaying and erasing menu and image data, makes visual interface soft for human eyes. The system we have developed is a typical example of multimedia applications. We expect the image workstation will play an important role as a platform for multimedia applications.

A technique is presented for intraframe color image data compression which produces visually lossless imagery compared to the original. This algorithm consists of a color vector quantizer operating in the Luv uniform color space, followed by a reversible codeword assignment strategy that uses prediction to achieve conditional entropy type bit rates. Unlike differential pulse code modulation (DPCM), predicted values are used for codebook selection instead of the computation and coding of a residual signal.

We built a multimedia document system for the SIGGRAPH Interactive Proceedings to demonstrate the potentials and challenges in using technology to capture better the essence of SIGGRAPH conferences. The prototype system uses the NeXT computer system to present textual, mathematical, illustrative, colorful, audio, video and animated material. Special attention was given to including tools for interactive manipulation of images included in typical SIGGRAPH papers.