Algorithm Research Of Classified Coefficients Detection And Vision Optimization Based On DCT Domain

With the rapid development of the information technology,the high-definition video is becoming more and more popular for people of all ages,and video encoding technology is also becoming increasingly important as well.H.264/AVC is a kind of video coding standard that jointly launched by the two International organizations VCEG and MPEG,nowadays become one of the most popular video standards.The H.264/AVC standard has absorbed the advantage of previous video standards,meanwhile it adopt a variety of advanced video coding technology,which makes it more efficient than the conventional video coding standards increased by at least more than doubled.Yet at the same time,it also brings a lot of coding complex computation.Therefore,on the condition of ensuring the quality of the video coding,optimize the performance and reduces coding complexity for H.264/AVC video coding algorithm has an important practical significance and application value.This paper focuses on H.264/AVC standard DCT domain coding algorithm,mainly to do the research work as followings:(1)There are always a large number of zero-quantized coefficients(ZQDCT)obtained from motion-compensated 4×4 residual block after integer discrete cosine transform(DCT)and quantization process.To solve this problem and reduce the redundancy calculation,classified detection algorithm of zero-quantized DCT coefficients is proposed,that based on analysis and extract the best all-zero block detection threshold,and combine with Gaussian model and "Z" shape characteristic of DCT energy distribution.In the algorithm,the residual block coefficients can be detected as ZQDCT before DCT transformation according to the classification model,which can skip the transform quantization operation.Experimental results show that,compared with the standard algorithm,the proposed algorithm reduces the encoding data bit rate obviously,and achieve a highest timesaving of 44% during the process of the discrete cosine transform(DCT),and quantization(Q),inverse quantization(IQ)and inverse discrete cosine transform(IDCT).(2)Combined with the human visual system,a kind of just noticeable distortion(JND)model based on DCT domain is applied to the H.264 standard algorithm.As a result that human eye is a ultimate recipient of video image,which have greater sensitivity on the low-frequency DCT coefficients component,and high frequency portion may contain more visual redundancy coefficient.JND indicates the maximum image distortion that human eye cannot detect,reflecting the vision of redundant information in the image imperceptible.The proposed algorithm calculate JND thresholds corresponding the 4×4 DCT transform block,filter the DCT coefficients after the transform process.And then abandon the corresponding position coefficient which is smaller than the JND threshold value,thereby removing the subjective visual redundancy on video image.The following step is skip quantization operation to saving computation if the DCT coefficient is set to zero.Simulation results show that the JND model can save bit rate without affecting the quality of the video image on the video encoding,and achieve good results on visual redundancy optimization.Through the research of the H.264 standard on DCT and quantization process,by using classified Detection algorithm to predict ZQDCT coefficients in advance and skip the DCT encoding process,finally save the computation;on the other hand,considering the perspective of the human visual system,by analysising visual redundancy on DCT domain and then present algorithm to filter correlation coefficient redundancy and optimize the encoded video image quality.