Macro-block Mode Decision in MPEG-2 Video Compression Using Machine Learning

摘要

Video compression currently is dominated by engineering and fine-tuned heuristic methods. Presently storing and transmitting uncompressed raw video requires large storage space and network bandwidth so compression is required. Many compression algorithms proposed to solve this type of problem. In this paper, we design a machine learning approach for the video compression using MPEG-2 codec. Various video compression techniques encode the video frames by applying inter and intra coding scheme. Video frames are divided into macro-blocks and each macro-block is encoded either by inter or by intra coding technique. It is an important issue to decide which coding technique will be applied to compress a given macro block. To solve this problem, we applied the machine learning approach in MPEG-2 video compression. We used support vector machine for the learning process and after learning any macro-block can be classified in intra or inter coding. Our experimental result suggests that use of machine learning in macro-block mode decision in MPEG-2 increases the PSNR while preserves the encoding and decoding time.