Video coding optimization with visual-attention area extraction

Abstract

While statistical redundancy removal has been extensively studied as a core technology of traditional video encoders, visual attention can be further exploited to improve coding efficiency. In this paper, we present an optimization strategy of video coding based on visual-attention area extraction. The proposed algorithm is a three-step method. First, a visual-attention map is generated with a spatio-temporal attention model. Second, we utilize a morphological filter to get a binary mask, and then detect the visual-attention area in the video frame. Finally, to encode the visual attention areas with improved performance, we propose an optimization strategy based on the ABR RC method and mask. The experimental results show that the proposed method can achieve better subjective visual quality at given bit rates.