Usage of characteristic points to reduce the time of performance of SAD metric calculation during video flow compression by motion compensation method

Abstract

Listen

Translate article

Today, it is difficult to imagine a video system that does not use compression of the input video stream for transmission or storage, because the uncompressed size of video files can reach several hundred gigabytes. There are many different video stream compression techniques used in various codecs, however, one of the most commonly used is the motion compensation technique, which allows you to transmit video stream frames in the form of compensated inter frame difference, by finding the motion vectors of individual blocks image. Despite the fact that the above-mentioned technique has been used for the last twenty years, it still has two problem-priority places - assessments of the similarity of image blocks and the algorithm for searching blocks. This article focuses on the problem of estimating the similarity of the image, in order to reduce the time of estimating the similarity of the blocks, which in modern algorithms takes from 40 to 80% of the time of the entire process of encoding the video stream. The article considers the usage of metrics for estimating the similarity of images and their importance for the process of motion compensation during video stream compression, such as SSD (standard deviation) and SAD (sum of absolute differences). The temporal contribution of metrics to the video stream encoding process, in particular to the motion compensation process, is estimated. An algorithm combining classical SAD and parameter estimation based on characteristic points is proposed, which will reduce the metric calculation time for estimating block similarity. The reduction of SAD calculation time due to reduction of comparison points is estimated for three proposed comparison templates - HSAD (half SAD), TSAD (third SAD), DSAD (diagonal SAD). For the selected templates, the results of processing the test video sequence were compared with the results obtained during the processing of the video sequence by the classic SAD. The main attention was paid to the assessment of the following parameters: the relative difference SAD, the increase in the number of candidate blocks, the overlap of candidate blocks.