A Fast and Accurate Moving Object Extraction Scheme in the MPEG Compressed Domain

Abstract

The hot topic of moving object extraction has been transferred from the pixel domain to directly from the compressed stream. The existed work of compressed-domain moving object extraction in the literature usually can meet the requirement of real-time performance. However, the accuracy of extracted objects is too coarse. The reason is that the information utilized in the compressed-domain extraction process is not abundant and noisy. In order to solve this problem, a novel moving object extraction scheme with high accuracy for MPEG stream is proposed. Firstly, the contour-feature and motion vector based projection are fused to extract the coarse object in an efficient way. Second, the blocks in I frame are partially decoded to improve the segmentation accuracy. The innovation lies in a threshold method of image segmentation with automatic seeding is applied on the DC+2AC image of I frame to extract the contour-feature, which is combined with the projection of motion vector field of P frame to complete the coarse segmentation. Then tradeoff is made between the executive time and results' accuracy. High accuracy is acquired at the cost of partially decoding time and region-growing processing time, which is worthwhile at the expense of such a little more time to refine the result. Experiments reveal that the proposed algorithm has the attributes of both real-time and high accuracy at the same time.