Saliency region extraction for MPEG video method based on visual selective attention

Saliency region extraction is a key video processing technology. From the human visual selective attention (VSA) in perspective, a new Saliency region extraction approach in MPEG compressed domain is proposed. First, visual saliency features in the MPEG compressed domain are extracted. There are 4 types of junior static features specific to brightness, color, direction, texture and 2 types of junior dynamic features specific to magnitude and consistency, which are extracted from the discrete cosine transform (DCT) coefficients and motion vectors (MV). Second, Combined with the Itti´s VSA model framework, using Feature Maps, Conspicuity Maps and Saliency Maps, a saliency region is extracted successfully. Experimental results show that the extracted saliency region is consistent with visual perception, meanwhile the computation time of the method is able to meet the real-time requirements.