Research On Visual Based Efficient Video Coding Technology

With the rapid development of mobile communication technology and unmanned aerial vehicle(UAV)technology,multimedia network technology with digital video as the main content is no longer only for the traditional TV system.Especially,with the more and more prominent role that UAV plays in natural disaster monitoring,commercial performance,military assistance and other fields,the requirements of digital video coding technology are getting higher and higher.Although the traditional video coding technology has made great progress in the elimination of space redundancy,time redundancy and information entropy redundancy,it has no obvious results in eliminating visual redundancy.In the final analysis,the human eye is the final recipient of the video signal.Therefore,this paper proposes a coding system which supports three video streams simultaneously transmit for inadequate network bandwidth when single watch video by studying the characteristics of human vision system and drawing on the idea of multi-description coding.At last,through the experimental data it is found that the coding system of this paper have better visual effects than traditional H.264-based coding system in the case of similar bit rate and inadequate bit rate.In the case of similar visual effects,the system of this paper can save about 20% of the bit rate.This thesis studies a variety of bottom-up salient map models.According to the merit and demerit of each salient map model and the needs of application scene of the video coding system,a salient map model based on frequency tuning algorithm is selected,and an equalize image brightness and image chromaticity salient map model is proposed on the basis of the model mentioned previously.Experimental results show that the improved model is more effective to detect salient maps.Finally,this thesis achieves gain of the region of interest by using the improved salient map model.The just noticeable distortion(JND)model uses a quantized threshold to represent visual perception of redundancy.Human eye is unable to feel changes that are not above this threshold.Thus,any unnoticed information difference is not encoded into the video stream.This thesis separately studies contrast masking effect and background brightness masking effect and time domain masking of JND model,and in the final system to achieve the model.The visual attention model mainly exploits the highly uneven distribution feature of the visual cones on the retina.The distribution density of cells reaches maximum on the central fovea of retina.As the distance to the central fovea increases,the cell distribution density reduces very quickly.This causes that the human eye system gets the highest spatial resolution of in the visual center and the spatial resolution decreases rapidly as the distance from the image to the visual center increases.On the basis of the visual attention model,this thesis combines with the JND model,achieve the visual attention model based on the image content,and realizes the dynamic transfer of the visual attention by using the motion vector.