Dibrivny OA The method of increasing the efficiency of video stream transmission during compression by the method of motion compensation. - Qualifying scientific work on the rights of the manuscript. Dissertation for the degree of Doctor of Philosophy in specialty 123 - "Computer Engineering". - State University of Telecommunications. - Kyiv, 2021.The dissertation is devoted to the research of motion compensation technique, which is used in most modern technologies of video stream compression and based on frame transmission in the form of compensated interframe difference, in order to reduce compression time of video stream. The current state and prospects of development of video file compression technologies are analyzed, the main algorithms and techniques that are part of them are determined. The main problem areas of such technologies are identified. The development of the resulting methodology took place by eliminating the problems of modern techniques. Based on the analysis of the scientific literature, the main metrics used to assess the similarity degree of images were identified. It was found that in modern video compression systems as the main metric used SAD, the calculation of which, for all its simplicity and low computational complexity can take from 40 to 80% of the total coding time of the video stream. To eliminate this shortcoming, it was proposed to use comparison templates based on characteristic points, which allow to reduce the calculation of the metric by reducing the number of comparison points. When evaluating the performance of the algorithm for estimating the similarity of images based on characteristic points, it was determined that reducing the number of control points leads to a sharp decrease in PSNR, due to loss of information in areas with high image entropy and the emergence of persistent uncertainty in areas with low image entropy. To eliminate the loss of information in areas of high entropy, it was proposed to use algorithms for selecting the edges of the image and refining the candidate blocks on the map of the edges of the image. The convenience of this approach is due to the fact that the process of finding motion vectors is for the video stream in YUV format, namely for the luminance component of the image (Y) and partial ignoring of color components. Y is essentially a grayscale image. This approach has significantly reduced the loss of PSNR with relatively small reductions in the speed of calculation of motion vectors. The next step in the formation of the resulting technique is to improve existing methods of block search, to eliminate the main disadvantages of using a template comparison. The formation of the block search method for the resulting technique was carried out by improving the four main parts of block search algorithms: 1. Pre-processing of the video frame, by calculating the uncompensated difference with the previous frame, and zeroing the motion vectors of the 8 blocks, the maximum value of which does not exceed a specified threshold, to eliminate ambiguity in low entropy areas. 2. Formation of an extended set of more accurate predicates, to reduce the time of refinement of candidate blocks and to exclude the influence of random coincidences of pixel values on the formation of the motion vector. 3. Extension of the conditions for the formation of threshold values for the conditions of early search. 4. Forming a set of adaptive templates to specify the blocks from which the motion vectors will be formed. The main approach for the formation of these four parts of the method was to use adjacent blocks not only of the current and previous frames, but also the next frame, by pre-processing part of the blocks of all frames of the video sequence. This approach is due to the constant movement of objects over short periods of time. As a result of a combination of template comparison for estimating block similarity, edge selection algorithms and an improved block search method, the resulting method of finding motion vectors. Both modifications of the method were tested according to the comprehensive testing recommendations of the international commission MPEG for the evaluation of complex video tools, on the example of 13 video sequences, with different bitrates. The effect of increasing the size of the blocks on which the 9 image is divided on the PSNR value and the maximum acceleration was also investigated. Comparison of the results of the proposed method was carried out with modern methods of motion vectors: FS, TSS, NTSS, DS, MVFAST, PMVFAST, which determined that the proposed method allows to obtain acceleration at 15-17% from the nearest analogue at losses at the level of 0.3-1.2% of PSNR values from the values of the reference vector search method motion for modification A and acceleration at the level of 27-31% with a loss of 1.1% -2.1% depending on the block size and bitrate of the video sequence, for modification S.
