Research On Rate Allocation Based On Region-of-saliency

With the rapid development of the Internet and mobile communications, video telephony, video conferencing, wired / wireless video monitoring become increasingly-common in people’s lives. The requirements of clarity of the video images, real-time are also increasing. The Video Compression Coding technology demands move forward rapidly since the video information data which withstand within a limited bandwidth are increasing continuously. The rational allocation of bit rate is the key of research in the field of video coding. However, the reconstructed image can not achieve optimal subjective visual effect under limited rate resources when conventional video coding techniques allocate the same quantization parameter for each macroblock in one frame image. Focuing on this shortcoming, this paper research extract the saliency region of video image and allocate bit rate according to the difference of saliency image to achieve a better image quality comparing with conventional distribution method based on visual characteristics of human visual system.Research works mainly includes the following four aspects:(1)Proposed a temporal saliency map generation algorithm based on the relative motion information of the video image. Firstly, use the pyramid optical flow and eight parameters perspective model to obtained the relative motion vectors of two adjacent frames; then use the relative motion vectors to obtained the temporal saliency maps of video image.(2)Proposed a multi-feature maps fusing method based on the temporal saliency maps of video image. Firstly, use color contrast and texture information to calculate the spatial saliency maps. Secondly, perform threshold processing to spatial and temporal saliency maps respectively. Finally, get spatial-temporal saliency maps by fusing spatial and temporal maps according to linear proportion.(3)Proposed a rate distribution and coding algorithm based on the macroblock layer of HVS saliency region, this algorithm, according to the saliency map, assess the significance degree of each macroblock in a whole frame, and then assign different quantization parameters to each macroblock to achieve rational distribution of bite rate.(4)Proposed an improved PSNR. Considering visual characteristics of the human visuals system, improve the weight of saliency region when calculating PSNR. Use the improved HPSNR to assess the quality of the encoded images.In this paper, we test the coding algorithm by large amounts of video images. The experimental results indicated that the proposed algorithm can effectively extract saliency region of the video images, and allocate at a rational ratio in the situation of quantization parameter adjustment guided by saliency region. Using this algorithm can achieve a better encoded image quality consistent with the human visual characteristics than conventional video coding.