Video Object Segmentation In H.264/AVC Compressed Domain

Since most videos are converted into the compressed format, the video object segmentation in the compressed domain attracts extensive attention in recent years. The video object can be extracted in the compressed domain where the video need not be decompressed completely. The motion information and coded residual extracted from the compressed bits stream with entropy decoding are used as the motion characteristic and texture feature. Contrast to the detection algorithm in pixel domain, the computation load and time decrease and it can meets with the real time requirement. At present the detection algorithms in compressed domain focus on MPEG compressed domain and has less study in H.264/AVC domain. As the up to date compressed standard, the coding efficiency of H.264/AVC standard is twice as the MPEG standard many specialities can adapt to network. The research on video object detection in H.264/AVC compressed domain has the academic meaning on video broadcasting and communicating,multimedia services.This paper analyzes the information included in compressed bits stream, and has a research on motion detection algorithm in H.264/AVC domain based on subsistent detection algorithm in compressed domain. Research includes the following aspects:(1) Aim at video with moving background, the paper estimates the global motion model parameters based on RANSAC algorithm to get the compensated residual motion field; and in view of the dectction result with MV is rough, the paper improves the detection reracity combining the DCT coefficient of adjacent frame.(2) Based on the analysis of threshold selecting on entropy, the coded bits of the Gaussian noise after DCT transform similar Gaussian distribution, the paper gets the better results by the local self-adaptive threshold.