Automatic extraction of moving objects in video sequences based on spatio-temporal information

The emerging video coding standard MPEG4 describes video sequences as video objects before coding, which enables various context-based functionalities for multimedia applications and improves the coding efficiency as well. Extraction of moving objects in video sequences is a significant technology for implementing the emerging object-based video coding standard MPEG-4. Automatic extraction method has been main development trend in comparison with semi-automatic method in that it needs no manual intervention and can reach the real-time requirement However, to extract moving objects in video sequences automatically and exactly is not an easy business. In this paper, an algorithm for automatically extracting video moving objects based on spatio-temporal information is proposed. At first, initial binary moving mask representing moving regions is obtained by hi-h-order statistics model based on temporal motion information. Then, Markov Random Field (MRF) classification model is established based on image primitives to attain the more complete and credible binary moving mask. Afterwards, an improved watershed algorithm based on non-linear transform is developed to segment moving regions processed by morphological opening and closing. Finally, moving objects are extracted by employment of ratio method on spatial and temporal results. Satisfying experimental results are achieved, which illustrates the efficiency of proposed algorithm.