Selective H.264 Video Coding Based on a Saliency Map

The demand in modern multimedia data transmission is continually increasing. New compression standard, such as the recent H.264/MPEG-4 AVC video coding standard, drastically improves the compression ratio. This higher compression ratio is required because the amount of multimedia data to transmit increases and that the perceived quality expected by the end user is not lessened. A complementary solution to maintain or to improve the perceived video quality is to selectively compress the information: perceptually non-important areas are subjected to higher compression than more relevant parts of the data. Therefore, the perceived quality can be heighten without increasing the bit rate. In this paper, we propose a new two-pass based coding framework driven by a saliency map. A saliency map is a 2D topographic map encoding the capacity of each pixel to attract the visual attention. This map stems from a computational model simulating the human visual attention. A selective compression is achieved by an a-priori knowledge of the spatial locations of the visually important areas. First, the proposed algorithm determines a quantization step for each macroblock according to a given bit rate (inferior to the global transmission rate). The second pass allocates the remaining bit budget by using the saliency map. Experiments based on the H.264 coding scheme were conducted leading to an undeniably objective quality improvement of the most interesting areas.