This paper presents a reconfigurable machine for applications in image or video compression. The machine can be used stand alone or as a universal accelerator co-processor for desktop computers for image processing. It is well suited for image compression algorithms such as JPEG for still pictures or for encoding MPEG movies. It provides a much cheaper and more flexible hardware platform than special image compression ASICs and it can substantially accelerate desktop computing.

Pyramid codes such as 3-D subband coding, wavelet and fractile can be implemented with much simpler encoders and decoders than DCT-based compression systems. Since they do not use blocks, they do not suffer from blocking artifacts. They are particularly useful when the cost of the encoder is a concern, such as picture phone, teleconferencing, and camcorders. If used in production, post-production processing can be implemented without degradation. A programmable real-time encoder and decoder have been constructed for evaluating color 3-D subband coding algorithms. The results of tests with this hardware are presented for a wide range of compression ratios as well as the post-production processing of compressed video.

In this work we develop an adaptive scheme for quantization of subband or transform coded frames in a typical video sequence coder. Using a simple statistical model for the subband or transform coefficients, we present a procedure to determine the optimum dead-zone quantizer for a given entropy of the quantizer output symbols. We find that, at low bit rates, the dead- zone quantizer offers better performance than the uniform quantizer. The model is used to develop an adaptive procedure to update the quantizer parameters on the basis of the state of a channel buffer with constant output rate and variable input rate. Experimental results show that the model can be effectively used in a practical scheme for buffer control.

A solution for real-time compression of digital YC_RC_B video data to an MPEG-1 video data stream has been developed. As an additional option, motion JPEG and video telephone streams (H.261) can be generated. For MPEG-1, up to two bidirectional predicted images are supported. The required computational power for motion estimation and DCT/IDCT, memory size and memory bandwidth have been the main challenges. The design uses fast-page-mode memory accesses and requires only one single 80 ns EDO-DRAM with 256 X 16 organization for video encoding. This can be achieved only by using adequate access and coding strategies. The architecture consists of an input processing and filter unit, a memory interface, a motion estimation unit, a motion compensation unit, a DCT unit, a quantization control, a VLC unit and a bus interface. For using the available memory bandwidth by the processing tasks, a fixed schedule for memory accesses has been applied, that can be interrupted for asynchronous events. The motion estimation unit implements a highly sophisticated hierarchical search strategy based on block matching. The DCT unit uses a separated fast-DCT flowgraph realized by a switchable hardware unit for both DCT and IDCT operation. By appropriate multiplexing, only one multiplier is required for: DCT, quantization, inverse quantization, and IDCT. The VLC unit generates the video-stream up to the video sequence layer and is directly coupled with an intelligent bus-interface. Thus, the assembly of video, audio and system data can easily be performed by the host computer. Having a relatively low complexity and only small requirements for DRAM circuits, the developed solution can be applied to low-cost encoding products for consumer electronics.

The central payload video system, CPVS, is a highly flexible central video facility designed to perform processing on payload video as well as system video. Although developed under the Columbus program the CPVS can be easily integrated into any space station program such as Freedom and MIR. The CPVS is compatible with component video (Y,C) as well as with composite video (standard NTSC and PAL). The CPVS also supports non standard video high resolution (geometrical) and high frame rate (excellent time resolution) video. The CPVS is converting all received video into digital video. The standard formats are compatible with the world wide recognized standard, D1 (CCIR 601 and CCIR 656 compatible), and the non standard formats are compatible with an extrapolation of the D1 standard. The CPVS is able to process up to eight (8) simultaneous video channels. The CPVS is able to route any input to any or all processing elements using an internal switching and routing unit. The number and type of processing elements can be chosen according to every mission needs and new processing standards can be supported with little or no modification. The selected D1 standard ensures excellent video quality, and the fully redundant design of the CPVS ensures maximum availability of the system for the users. The selection of standard video formats and standard processing algorithms (compression) allows the use on ground of low-cost, off-the-shelf video equipment. The CPVS operation is based on tables which are generated on ground and loaded on board to be executed at a given on board time without the need of real time interaction.

Telepathology is aiming at pathological diagnoses based on microscopic images of cell samples through broadband networks. The number of pixels in pathological microscopic (PM) images is said to be approximately 4 to 6 million. In this paper, digital PM images are made without films using a super high definition (SHD) image prototype system, which has more than double the number of pixels and frame frequency than those of HDTV images. First, color distribution and a spatial spectrum are analyzed in order to estimate compression characteristics of the images. In addition, the lossless and lossy JPEG coding characteristics are investigated. In the lossless compression, the PM images have compression ratios which are very close to 1, while the general images have compression ratios around 2. The PM image compression ratios in the lossy JPEG coding, where the L*a*b* color difference is less than 2 to 3, are found to almost equal those of the lossless JPEG (Joint Photographic Coding Experts Group) using arithmetic coding. The PM image coding performance in the lossy JPEG coding is also found to be inferior to that of general images including still life images, portraits, and landscapes.

Component-based color video signals usually consist of one luminance (Y) and two chrominance or color-difference (U and V) components, which are obtained by multiplying the R, G and B components produced by the video camera by a 3 X 3 matrix. In order to compensate for the nonlinearity of the TV monitor on which the video signal will be displayed, the R, G, and B signals are usually first gamma-corrected before the matrix operation is applied. Due to this gamma correction, the Y component does not represent exactly the real luminance L of the recorded scene, and part of the real luminance information is carried by the U and V components. The introduction of errors into these chrominance components by, for example, video coding will therefore lead to perceivable errors in the luminance produced by the TV monitor on which the coded signal is eventually displayed. In this paper, we present a simple but effective method for avoiding this crosstalk of chrominance errors into the luminance. This method can be incorporated in most compression systems.

The large variety of algorithms for data compression has created a growing need for methods to judge (new) compression algorithms. The results of several subjective experiments illustrate that numerical category scaling techniques provide an efficient and valid way not only to obtain compression ratio versus quality curves that characterize coder performance over a broad range of compression ratios, but also to assess perceived image quality in a much smaller range (e.g. close to threshold level). Our first object is to discuss a number of simple techniques that can be used to assess perceived image quality. We show how to analyze data obtained from numerical category scaling experiments and how to set up such experiments. Second, we demonstrate that the results from a numerical scaling experiment depend on the specific nature of the subject's task in combination with the nature of the images to be judged. As results from subjective scaling experiments depend on many factors, we conclude that one should be very careful in selecting an appropriate assessment technique.

In the foreseeable future the application of digital techniques in production, transmission, storage and processing as well as in distribution of television signals will involve the coexistence of various signal formats including analog ones and the increasing use of bit rate reduction procedures. The possible impairment of the picture quality by the application of various and possibly cascaded data rate reduction procedures in the framework of this development has not yet been clarified sufficiently. In order to support the public broadcasters of Germany, Austria, and Switzerland in introducing and utilizing digital techniques, the IRT carried out studies on this question in close cooperation with the broadcasters. For this purpose, a scenario of all relevant television chains from production to transmission was drawn up, a source tape including a selection of suitable test sequences was produced and these sequences were fed in all the selected chains. The results were recorded and subjective tests were carried out to evaluate the impairment of the picture quality. This paper includes the results of a quality assessment by experts at an intermediate stage of the project. The results of subjective tests are presented.

The paper addresses the problem of reducing the computational load in block-based motion estimation (BBME) by exploiting a preliminary motion field, which is obtained from the coarsest level of a multiresolution pyramid, and its successive refinements in the finer ones. The first estimation is achieved by a full search, even on greatly reduced scales, whereas at the other levels just a small correction is computed, thereby achieving a speed increase. However, this results also in difficulty to improve an inaccurate initial estimation. The algorithm proposed in the paper overcomes such a drawback and even outperforms BBME in terms of rate vs. distortion through the computing of a variable-resolution motion field which allows the same reconstruction quality with fewer vectors. A good correction of wrong initial estimates is achieved by use of an appropriate propagation of displacement vectors from one level to the next one and of an adaptive search range. The vectors propagation is carried out by choosing the best vector among a set of neighbors taken from the previous level and among the already refined ones in the current level. The search range is chosen on the basis of the error in reconstructing the block related to the propagated vector. Finally, the adaptive resolution of the field is achieved by ending the propagation at an intermediate level of resolution, if a vector allows a good reconstruction quality of its block.

A region-based multi-criterion segmentation algorithm for image sequence coding is described in this paper. It is based on morphological segmentation, motion estimation, and compensation techniques. The concept of multi-criterion is introduced in the traditional gray-level segmentation process, which takes into account the high temporal redundancy. The current image is segmented into two different kinds of regions: homogeneous gray-level regions coded by texture modeling method or homogeneous motion regions represented by motion compensation. No prediction error needs to be transmitted. Contours are simplified by a new non-linear filter. Preliminary experimental results have demonstrated high potential performance, with the combination of the power of contour-texture approach and that of the motion compensation technique, for very low bitrate image sequence coding.

In many video coding schemes, especially at low bitrates, a temporal subsampling of the input image sequence is considered. It is realized by skipping images at the transmitter, and these frames must be reconstructed at the receiver side. In order to prevent jerkiness or blurring in the moving areas, motion-compensated interpolation has to be applied. Frame interpolation algorithms usually consider two consecutively transmitted frames and assume constant velocity linear motion between these frames. The algorithm proposed in our paper assumes constant acceleration -- instead of constant speed -- that is estimated based on three consecutively transmitted images and two corresponding motion fields. This approach is much closer to the real situation than the previous methods taking into account only constant speed motion. The proposed interpolation algorithm results in smoother movement in the final interpolated video sequence. Experiments carried out using real-life image sequences confirm the applicability of the proposed method for low and very low bitrate video coding.

The paper investigates the capabilities of a novel 2D motion estimation and object-background segmentation algorithm to be used for low bit rate videophone compression. The algorithm is based on a model of the image inspired by the work of Geman and Geman [GG84]. Observable are three frames, the previous, the current, and the successive ones, and the edges of the current frame. The algorithm tries to minimize a global cost function, depending on local displacement and segmentation values, whose minimum corresponds to the desired displacement and segmentation fields. The performances of the algorithm have been measured on the total residual error and on the ratio of the final to initial costs. The algorithm has been tested on artificial and natural image sequences and the preliminary experimental results have been satisfactory, confirming the validity of the model and showing a perceivable improvement over block matching. Moreover the algorithm is suitable for distributed parallel processing.

In many codecs a predicted image is calculated by motion compensation on a block-basis. One motion vector for each block is determined and the prediction used is a displaced block from the previous image. This method has one major disadvantage: blocking artifacts. In this text we investigate the use of an interpolated motion field to solve this problem. A discrete number of motion vectors is used to characterize the motion field. With these vectors a motion vector for each pixel is calculated. If we use this motion field for prediction all the blocking artifacts disappear thus giving a visually better prediction. Using the motion vectors resulting from the block-based codec, which are not optimized for this interpolation, and interpolating them has one drawback. It decreases (compared to blockmatching) the signal to noise ratio of the prediction, leaving a bigger error to code, and thus the gain of a visually good prediction is partially lost. We propose an algorithm for the calculation of motion vectors that solves this problem. Knowing the fact that we will interpolate, we are able to guess the motion field on a pixel basis for each possible motion vector. Thus we are able to get an optimal motion vector for this guessed pixel based motion field which is very close to the resulting pixel based motion field. By interpolating these motion vectors we are able to increase both the subjective and objective quality of the predicted image.

To reduce temporal redundancies appearing in video sequences block based motion estimation techniques have been successfully applied. However, due to the block-based model blocking artifacts appear in the resulting displaced frame difference reducing the global coding scheme efficiency. In this paper, a motion field refinement technique is proposed to overcome these artifacts. The method is based on segmentation and vector quantization (VQ). The motion field resulting from the block-based motion estimation is segmented using a set of patterns stored in a codebook. These patterns are derived by the LBG algorithm trained on natural edges. The VQ technique allows a low cost data transmission of the segmentation information. Finally, the segmentation is controlled by an entropy criterion, which performs an optimal bit allocation between motion, segmentation, and DFD information. Simulation results show that a significant improvement is achieved using the proposed method. In particular moving edges are much better motion compensated and the visual quality is greatly enhanced.

An efficient video coding algorithm is described. It is based on a new quadtree block merging algorithm and wavelet decomposition of video signals using orthonormal and biorthogonal filter banks. The merging algorithm represents the motion compensated predicted images in a small set of rectangular regions in order to improve the coding efficiency and to minimize the border affects associated to the wavelet transform. Results of coding simulations using 8 X 8 and 16 X 16 blocks are presented. Also, subjects such as choice of the optimal filter bank, boundary effects, quantization of the wavelet coefficients, and filtering the prediction error by wavelet decomposition are discussed.

A motion-compensated sub-band coding (SBC) scheme for video signals, featuring fixed-rate and optimum quantizer, is presented. Block matching algorithm provides a suitable inter-frame prediction, and a 64 sub-band decomposition allows a high decorrelation of the motion- compensated difference field. The main drawback is that sub-bands containing sparse data of different statistics are produced, thus requiring run-length (RL) and variable length coding (VLC) for best performance. However, most digital communication channels operate at constant bit-rate (BR); hence, fixed-rate video coding is the main goal, in order to reduce buffering delays. The approach followed in this work is modeling the subbands as independent memoryless sources with generalized Gaussian PDFs and designing optimum uniform quantizers with the goal of minimizing distortion after a BR value, also accounting for the entropy of the RLs of zero/nonzero coefficients, has been specified. The problem is stated in terms of entropy allocation among sub-bands minimizing the overall distortion, analogously to optimal distortion allocation when fixed quality is requested. The constrained minimum is found by means of Lagrange multipliers, once the parametric PDFs have been assessed from true TV sequences. This procedure provides the optimum step for uniform quantization of each sub-band, thus leading to discarding some of the least significant ones.

An analysis of the motion estimation system for the MPEG-2 video coding algorithm is reported. With the target of single-chip VLSI implementation, not only the number of operations but mainly the cache size and the memory bandwidth are taken into account as feasibility issues. A multi-minimum motion estimation routine is described but, even if the complexity requirement is strongly reduced, it is not enough to allow large search areas. Finally, a vector tracing approach is proposed to overcome the analyzed drawbacks.

The MPEG-2 standard includes a spatial scalable hierarchy. For graceful degradation purposes, the base layer is a downscaled version of the enhancement layer. For this case of spatial scalability weighting matrices and spatial and temporal prediction modes are examined, resulting in an optimization of the enhancement layer encoder also with respect to hardware complexity.

A new coding scheme for combined natural and computer rendered image sequences is presented. It is specifically suited for the heterogeneous data sets used by an electronic set system at the receiver site for the synthesis and mixture of transmitted image sequences. The different types of data sets and their particular properties regarding data compression are explained. Finally, results are given comparing the new coding scheme with traditional MPEG2 coding based on typical test sequences.

An image coding system based on an interactive object oriented language which defines a structure of features is described. Features of color pictures are analyzed. Our system is designed using these features to paint a component picture and a composed picture with many components. Concerning the system, the image modification method is obtained to generate various kinds of reproduced pictures. The properties of motion are classified. Based on the motion classification and the image modification, moving pictures are produced. In order to generate satisfactory pictures efficiently, the function of interaction between client and server is provided. The interactive scheme of the feature coding is explained. The system is implemented by the language, Extensible WELL (Window-based Elaboration Language).

In this paper, a low computational complexity and high efficiency algorithm for progressive image compression is presented. The image is partitioned into blocks of pixels and a prime- factor 2-D discrete cosine transform (DCT) is applied. Because of the regular structure the DCT algorithm, the coefficients could be computed in groups. Each group of DCT coefficients is then vector quantized by using the pyramid vector quantizer algorithm. Since, the DCT coefficients of each group are allocated with a different number of bits, three PVQs are used with different radii.

Applications involving transmission of color images through low bandwidth telecommunication media involve a trade-off between image quality and decoding complexity. In this paper we investigate techniques that locally optimize the output quality of foreground objects in the decoded image, while obtaining significant gain by coarser coding of information of low interest areas of the image. High compression/quality ratios are achieved by introducing high-quality windows, as well as by using progressive coding principles in image filtering, and enhanced run length encoding techniques. The system described is currently fully operational and serving telemedicine image communication applications for remote diagnosis.

This paper describes how a generalization of the wavelet transform -- the multiresolution Fourier transform (MFT) -- can be used in model-based coding, in which image features including boundary contours and textures can be extracted directly from the transform coefficients. As such, it has the potential to extend transform coding to very low bit rate, feature-based compression. Results of the work are presented to demonstrate the effectiveness of the methods.

In this paper, a new vector quantization scheme for encoding images is presented. This proposed scheme is called multi-stage hierarchical address vector quantization (MSHAVQ) and it combines the concepts of the multi-stage VQ (MSVQ), of the hierarchical VQ (HVQ), and of the address VQ (AVQ). Its encoder consists of several stages where the hierarchical VQ (HVQ) or the address VQ (AVQ) are applied while the larger image blocks are divided into smaller sub-blocks. Digital decimation and interpolation techniques are used to reduce the blocking effect in the reconstructed images. The experimental results show that the bit rate of the last stages of the MSHAVQ are reduced by a factor of two or three when compared with a memoryless VQ system and that the proposed coding scheme yields to images having an SNR value about 35 db at bit rate around 0.9 bits per pixel.

This paper considers the role of video compression in the transmission of television signals. It highlights contribution and distribution as well as direct-to-home broadcasting applications on satellite which are current and describes aspects of the standards, technology and systems required to support them. Some aspects of system performance are discussed together with a brief summary of future developments particularly on other media than those traditionally used by broadcasters.

The European Union RACE (R&D in advanced communications technologies in Europe) and the future ACTS (advanced communications technologies and services) programs have been contributing and continue to contribute to world-wide developments in audio-visual services. The paper focuses on research progress in: (1) Image data compression. Several methods of image analysis leading to the use of encoders based on improved hybrid DCT-DPCM (MPEG or not), object oriented, hybrid region/waveform or knowledge-based coding methods are discussed. (2) Program production in the aspects of 3D imaging, data acquisition, virtual scene construction, pre-processing and sequence generation. (3) Interoperability and multimedia access systems. The diversity of material available and the introduction of interactive or near- interactive audio-visual services led to the development of prestandards for video-on-demand (VoD) and interworking of multimedia services storage systems and customer premises equipment.

This paper describes a rough hierarchy for an on-demand protocol design based on several concepts we believe would clarify the overwhelming complexity of multimedia application. These concepts include `semantic' control and `syntax' control, simple medium and structured medium, protocol switching, etc. A certain operation that may be performed on a medium is defined in terms of medium characteristics and dictated by semantic control. Where medium characteristics are ciphered in syntax control information. According to the complexity of these characteristics a hierarchy can be set up and protocol designs can be clearly classified, so that the concepts mentioned above are involved. Therefore, we can design protocols step by step, from basic ones to specific ones, and easily decide what functions should be realized in which set of protocols.

Specific mechanisms are required to carry MPEG coded information over an ATM network. After a brief description of the MPEG-2 video coding algorithm, this paper discusses the `isochronism' related requirements for the transmission of the coded data and the problems associated with the use of an anisochronous transmission mechanism. The main problems, concerning the choice of the ATM adaptation layer (AAL) for each level of interaction, are discussed. Even if the MPEG stream is inherently bursty, the technical literature considers both the constant bit rate (CBR) and variable bit rate (VBR) solutions. The paper reports on a study concerning the AAL choice, in relation with the degree of interactivity required by the application and with the delay constraints. A further issue is an overview of different scenarios for the provision of video services, by `MPEG or ATM.' Different possible network and service architectures are analyzed, and various solutions for storing video information are proposed. Three different service architectures are discussed: the fully centralized, the quasi- centralized, and the fully distributed. The last issue dealt with in the paper concerns a very attractive class of interactive video services, video-on-demand (VoD). In particular, the impact of the digital storage media-control command (DSM-CC) in the provision of a particular VoD service, the `video rental' service, is described. The network requirements to support DSM- CC commands allowing the user to perform the typical video cassette recorder (VCR) functions (freeze frame, fast forward, fast reverse) are stressed. Alternative approaches for the implementation of each command are discussed together with their impact on the traffic network.

In this paper, a video conference system we developed recently is presented. In this system the video codec is compatible with CCITT H.261, the audio codec is compatible with G.711 and G.722, the channel interface circuit is designed according to CCITT H.221. In this paper emphasis is given to the video codec, which is both flexible and robust. The video codec is based on LSI LOGIC Corporation's L64700 series video compression chipset. The main function blocks of H.261, such as DCT, motion estimation, VLC, VLD, are performed by this chipset, but the chipset is a nude chipset, no peripheral function, such as memory interface, is integrated into it, this results in great difficulty to implement the system. To implement the frame buffer controller, a DSP-TMS 320c25 and a group of GALs is used, SRAM is used as a current and previous frame buffer, the DSP is not only the controller of the frame buffer, it's also the controller of the whole video codec. Because of the use of the DSP, the architecture of the video codec is very flexible, many system parameters can be reconfigured for different applications. The architecture of the whole video codec is a streamline structure. In H.261, BCH(511,493) coding is recommended to work against random errors in transmission, but if burst error occurs, it causes serious result. To solve this problem, an interleaving method is used, that means the BCH code is interleaved before it's transmitted, in the receiver it is interleaved again and the bit stream is in the original order, but the error bits are distributed into several BCH words, and the BCH decoder is able to correct it. Considering that extreme conditions may occur, a function block is implemented which is somewhat like a watchdog, it assures that the receiver can recover from errors no matter what serious error occurs in transmission. In developing the video conference system, a new synchronization problem must be solved, the monitor on the receiver can't be easily synchronized with the camera on another side, a new method is described in detail which can solve this problem successfully.

In this paper, we design a new quantizer on studying the strategy for adaptive quantization proposed by MPEG-2 TM5. To exploit local perceptual characteristics in the macroblock, block-based visual activities are analyzed and then macroblock characteristic is identified by evaluating each block's local activity within it comprehensively to decide the adaptive perceptual quantization factor. Experiment results show that coding distortion is uniformly distributed across the image with the proposed quantizer, and picture quality is improved. In particular, block artifacts in flat area and distortion at edges on flat backgrounds are significantly reduced.

Since the approval of the H.261 standard many compatible video codecs have been announced. These hardware systems are not versatile since they are generally based on proprietary chip sets specifically designed for that application. In this paper we describe an implementation of a H.261 compatible video codec based on a digital signal processor (DSP). Since most of the coding tasks are performed by the DSP, the system can be easily programmed by the user to implement other algorithms. When the system is programmed for the H.261 algorithm, it is capable of encoding and decoding 15 Hz QCIF bit streams.

This document intends to describe the direction that future audio-visual standards should take. Within MPEG (ISO/IEC JTC1 SC29 WG11) discussions on this issue have led to the initiation of a new standardization activity called MPEG 4. A constantly changing world requires the development of new ways to come to standards for audio-visual applications.

In many parts of the world, telemedical services are being implemented for improving the quality of the health care system. Increasingly, however, they can also be seen as a potential for cost reduction of medical diagnostic and therapeutic activities. The technical basis for satisfying very demanding quality of service parameters in medical environments is made possible through networks with high bandwidths.

The Hospital Edouard Herriot in Lyon and 3M company, associated with the Electronic Department of Physics Chimics and Electronic Engineering School (CPE), decided in 1993 to begin a study on a project of image network. This project is composed of many practical applications to be checked one by one. The purpose of this paper is to discuss the context, which is kind of small picture archiving and communication system (PACS), to explain the methodology which has been used related to hardware and software implementation, and to give examples of the first results obtained. One of the main interests of the results is the possibility to obtain on the same support, 3M laser imager, a film including images from different modalities and abstract summing up the patient stay in the hospital. The framework used is built around Omnis7 and C++ language on a PC computer.

Current referral practice for ultrasonically detected fetal abnormalities contributes to parental anxiety, patient inconvenience, diagnostic inaccuracy, and service inefficiency. To determine whether telemedicine would reduce these disadvantages, we established an ISDN 30 link between a district general hospital on an island and a subspecialty referral center approximately 120 km away on mainland Britain. Live ultrasound images of the fetus were transmitted in real time from a commercial scanner in the hospital using a total data transfer rate of 2 Mbit/s. After decompression at the receiving end, there was almost no perceptible loss of picture quality or frame rate. This report describes the technical aspects of the link and discusses the implications for referral practice in fetal medicine.

This paper describes the HIM methodology and the results of this work reflected in a set-up for a pan-European ATM trial in health care. The work is aligned with the objectives of the Maastricht Treaty, which calls for the establishment of trans-European networks. The HIM methodology provides a structured way to come from user requirements to an actual implementation, providing along its way mechanisms and checkpoints which validate and support the final result. The user requirements are captured in user scenarios, which are then directed to specific implementation profiles. These profiles are obtained by mapping user requirements onto technical implementations. The underlying network infrastructure is envisaged to be the European ATM pilot, created by the European PNOs in alignment with their Memorandum of Understanding (MoU). Different access methods are used, which include access through MAN networks, through Ethernet and FDDI LANs, and workstations directly connected to the ATM network. The transport protocol used is TCP/IP. The results are described in situation and application specific contexts. The results and implementations are examined for their cost-effectiveness, and validated towards a broad user community.

The transmission of medical image data using asynchronous transfer mode (ATM) communication over a synchronous optical network (SONET) was evaluated. Key parameters and performance criteria of the ATM network were tested in order to assess the applicability of ATM technology to telemedicine. The ATM based network connection was provided by the local phone companies. The initial experiment involves two of three affiliated hospitals located within a 19 mile radius in California (the VA Medical Center, West Los Angeles; UCLA Medical Center; and Olive View-UCLA Medical Center). Each hospital has a picture archiving and communication system (PACS) within radiology. We designed an ATM over SONET network for PACS utilizing the TCP/IP communications protocol. We are implementing an ATM testbed over which we characterize the traffic and measure the total transmission time for large data files.

Two new techniques for the compression of 2D and 3D medical images are proposed in the paper. Both algorithms are based on DCT transforms and the aim is to ensure a nearly lossless image transmission and storage with a compression rate able to reduce significantly the amount of data. The first technique, used for the storage of 2D medical images (microscope, radiographic, x-ray images, ultrasonic images, computer tomograph sections) is based on the extraction of a `region of interest' in the original image. In fact, in most situations, only a limited region of the medical image is interesting for diagnosis. For compression, a direct DCT is applied on the original image, preserving a different number of coefficients inside and outside the region of interest (must be higher inside the region). The compression ratio obtained with this method depends on the size of the selected region and on the number of DCT coefficients preserved, ranging from 4 to 20. The second technique is based on the correlation existing between contiguous axial sections obtained from the CT. The 3D volume data is compressed using an original axial section and the difference images between successive sections. The advantage of this method consists in reducing the number of DCT coefficients necessary for a nearly lossless compression of the difference images. The compression ratio obtained ranges between 6 nd 10, without significant losses in the image quality.

An integrated optical circuit (IOC) combining all the detection functions of a standard magneto-optical reading head (MO reading, focus and tracking control) is presented. The reading function is achieved by means of a patented interferometric circuit. For tracking control, the well-known push-pull method has been applied and adapted to integrated optics. In this case, the role of the IOC is just to separate the beam reflected back from the disc in two halves in order to compare their intensity. For focus control, several principles have been tested: standard ones adapted to integrated optics (`wax-wane' method or Foucault knife-edge) and an original method based on multimode interferences, taking benefits of integrated optics specificity. The implemented technology is based on a silicon substrate with a silicon nitride core between two silica cladding layers (Si/SiO₂/Si₃N₄/SiO₂). This technology is a low cost technology well adapted for mass production. The optical components of the circuits are made by standard contact photolithography and reactive ion etching. Several wafers with about 50 devices each, have been processed and characterized. In particular, the detection signals have been compared with the detection signals delivered simultaneously by a Philips-IBM 128 MB MO drive. This experiment demonstrates the feasibility of an integrated detection device for MO drive.

The inherently high storage densities available using magneto-optic (MO) recording systems can be further enhanced by application of partial response signaling. In order to achieve this, equalization of the read channel is required to shape the channel to give the desired class 1 partial response. A number of analog and digital (FIR) equalizer functions are derived, and their performances assessed in terms of improved SNR due to increased resolution of the signaling. It is concluded that FIR filtering is superior to the low pass analog filtering used. Given typical figures for noise levels within a MO drive, it is verified that it is possible to push the recording density of the system up to the maximum achievable with this technique and still retain reasonable soft error rates.

A low cost byte error rate testing system for 90 mm magneto-optic (MO) disks had been developed. In order to achieve low cost and to perform the testing under realistic operating conditions, commercial magneto-optic disk drives were used instead of a high-priced, specialized disk tester. An IBM compatible 486-DX33 PC was used as the controller. The design goal of the testing system is to find the best cost/performance system configuration. A model was developed to predict the testing system performance based on the DMA data rate, SCSI bus data rate, MO drive data rate, access time, processor speed, and the number of sectors, N, involved in each read/write operation. Three configurations were studied: (1) only one MO drive; (2) two MO drives are connected to PC through separate SCSI host adaptor; and (3) two MO drives connected to PC through one SCSI host adaptor in a daisy-chain manner. The testing time of a disk varies from 40 min. to 24 min. when N is varied from 20 to 109. These testing times include the 10 min. of formatting time which is independent of N. The average access time of the MO disk was found to depend on the number of sectors involved in each read/write operation. Also, the measured access time is always lower than the specified 40 msec. These two phenomena can be explained by the smaller distance that the optical head needed to travel during the byte error rate testing. Based on the measured access time, a refined model was used to predict the optimal cost/performance configuration. A system with 6 or 7 MO disk drives connected to a PC was found to have the best cost/performance ratio.

Stereolithography is a part of rapid prototyping technology. It is possible to build 3D models with all their complexity in a very short time. The most developed technique stereolithography is a layer manufacturing technique (LMT). The object is built layer by layer. It looks like a reverse tomography. This technique is appropriate to store 3D information slice by slice.

A compact free space optical system that consists of novel hybrid source and receiver arrays for chip-to-chip and board-to-board optical communication is described. The hybrid source and receiver structures consist of the electronic chip, optoelectronic array and microlens attached and aligned using flip-chip bonding. The architecture is a three layered structure. Hemispherical shaped refractive microlens made of glass collimate the light from the microlasers and focus onto the photodiodes. Analysis was carried out using Gaussian beam propagation models and paraxial ray approximation matrices to analyze the hybrid optoelectronic arrays. The chip-to-chip distance was computed as a function of the microlaser beam waist. Furthermore, Huygen-Fresnel scalar diffraction was utilized to compute the optical crosstalk. Finally, the reliability of the system was evaluated under temperature cycling and flip-chip bonding process variations, which showed that such a system is potentially robust and non-sensitive to process variations.

In this work the questions of applications of chalcogenide glasses (CG) and the structures on their base in the technology of optical disks are considered. In brief are described the properties of high resolution inorganic thin film structures CG-Ag and CG layers. The peculiarities of laser lithography on such resists under the influence of sharp focused laser irradiation with the objective of formation of master disks and the tracking guides of optical disks were investigated. It is shown that under laser exposure the local heating of the resist and also the photostructural transformations, activated by this heating, provide the narrowing of the lines of the resistive mask in comparison to the size of the exposure light spot. Using the As₂S₃ layers the minimal width of lines 0.17 micrometers was obtained, under the exposure wavelength 476 nm and laser spot halfwidth 1 micrometers . This enables us to decrease the period of the tracking guides with the purpose of increasing the density of information recording. The possibility is shown of the formation of the tracking guides structure with the period up to the 0.6 micrometers . The thermosensitive structure CG-In(Sn) is developed for the nonerasable information recording by the irradiation of the semiconductor laser. The recording is carried out due to thermostimulated interaction between CG layers and metal, that leads to the substantial changes of the reflection coefficient. The media is characterized by the acceptable sensitivity, high contrast, and sufficient storage time.

Various modes of optical data storage are regarded. Basic practical systems for optical data storage of 2-D and 3-D images are systematized into groups characterized by similar means of data capture, recording and presentation. Analogue, digital, and interference modes of data capture utilizing only one recording media, as well as updatable and rewriteable recording media, are surveyed. Classifications are given. Novel terminology is introduced. Novel approach to 3-D data presentation is formulated.