Several new methods for vector quantization of image sequence were developed in this paper. The 3-d gradient operator1 and the newly developed stochastic gradient operator was used to estimate the motion vector on the present frame in a pixel-bypixel basis. Two new methods of block motion estimation using the pixel motion vectors were studied. The methods either use the pixel motion vectors directly on the moving block or calculate the block motion vector from the pixel motion vectors. Then for both of the methods, the differences between the prediction block and the moving block are vector quantized. The gradual deterioration of the image frames due to the incorrect matching of the code vectors and the difference blocks could be suppressed with the stochastic gradient operator and a simple smoothing filter for the prediction values. Good quality images for video conferencing applications could be obtained with the bit rate in the neighborhood of 0.15.

Several edge-adaptive dual-mode predictive coding systems are proposed for coding monochrome video signals. These algorithms use the pseudo-Laplacian edge detector to switch between the textural coding and the edge coding. The one-dimensional Song DM coder and the two-dimensional weighted average prediction DM coder were used for the textural coding. The edge coding was done with either a normal two-dimensional Song DM coder, Graham's DPCM schemes, or a DPCM coder developed by the authors. Nine different combinations of textural and edge coding techniques were developed and simulated. The bit rates of the various schemes were compared, and the range varied between 2.28-3.46 bits/pixel without further compression of the edge information bits or the correction bits. Subjective tests were carried out on images reconstructed according to these schemes. The results show that most of the pictures obtained with the techniques in this paper are superior to the pictures coded by the Song mode normal two-dimensional DM coder alone.

Telesign is designed to offer a means of visual communication over a 56- or 64-kbit/s data network. The purpose is to supply a means of visual telecommunication among the members of the deaf community using sign language or lip reading. The system consists of an edge detector followed by digital compression coding to meet channel requirements. Psychometric experiments have shown the need for 25 frames/s with a minimum definition of 128 × 128 points. Edge-detection techniques are discussed from the viewpoint of an unusual quality criterion: the subjective resemblance of the contoured image to the original and, closely related to this, the intelligibility of the image sequence. A classification of edge detectors is given, based upon the position of the contour with respect to the dark/light contrast boundaries in the picture. The relationship of this classification to the quality criterion yields the definition of a new gradient display, called the "shifted gradient," with improved visual quality. The pseudo-Laplacian, well-suited to the problem, is analyzed. Various binary coding techniques are reviewed, and the results of compression given. A psychovisual experiment, performed in conjunction with the INJS (National Institute for the Education of Young Deaf), Paris, France, is described; the results validate the proposed approach. Guidelines are proposed for the design of a real-time device. The application of Telesign to closed-caption TV programs for the deaf viewer is also suggested.

The reject and error rates of a certain hierarchical decision structure are derived under assumptions of statistical independence among the members of the hierarchy. It is shown that high decision reliability can be obtained with much more relaxed requirements on the individual recognizers in the hierarchy.

The importance of solving the measurement problem in character recognition systems before or collaterally with the decision problem is stressed. Measurements themselves are decision processes. In the system discussed here the measurements constitute decisions on the probable presence and most likely orientation of the edges of the character strokes in each elementary scanning area. This permits an effective and economical sequential feature detection in which the statistical variations due to printing and scanning variations are taken into account. The article concludes with a discussion of the role of the computer in recognition-logic design.