Perceptual Quality Optimization Of Hierarchical Prediction Video Coding

Hierarchical prediction structure has been widely used in video coding,video communication,video on demand as well as other fields due to its high coding efficiency.In H.264/AVC and H.265/HEVC standards,the typical implementations of the hierarchical prediction structure are hierarchical B-pictures structure and random-access structure,respectively.Focusing on the hierarchical B-pictures structure and random-access structure in video coding,this dissertation firstly studies and analyzes an important temporal noise,i.e.,temporal pumping artifacts.The causes and perception characteristics of temporal pumping artifact are analyzed and summarized.Then based on the related perception characteristics,the algorithm of estimating the regions with perceptible temporal pumping artifact is proposed.And for the hierarchical B-pictures structure,a quantization parameter cascading(QPC)method to suppress the temporal pumping artifact is also proposed based on the algorithm of estimating the regions with perceptible temporal pumping artifact.Furthermore,a QPC method based on the Lagrange rate-distortion optimization for the random-access structure is proposed.And then an efficient frame level rate control method based on the proposed QPC method is also proposed for the random-access structure.The main innovations and contributions are concluded as follows:1.Video subjective experiments are designed for analyzing the subjective distortions in the videos coded using hierarchical prediction structure.Firstly,based on the experimental observations,at low bitrate,video coding under the hierarchical prediction structure easily induces temporal pumping artifact.Then,the causes and important perception characteristics of the TPA are analyzed and summarized for the first time.These results are the basis for the subsequent researches about the temporal pumping artifact.2.An algorithm to estimate the regions with perceptible temporal pumping artifact is proposed for the first time based on related perception characteristics of the temporal pumping artifact and the visual masking artifacts of human eyes.Based on the experimental observations,texture and motion characteristics are the important factors affecting the perception of temporal pumping artifact.Therefore,the perception sensitivity of human eyes to the temporal pumping artifact is modeled based on the spatial-temporal masking and the texture and motion characteristics of the videos.Combining the established perception model and the fovea masking and the orientation masking,the regions with perceptible temporal pumping artifact can be estimated.Experimental results show that the estimation by the proposed algorithm is accurate and in line with human perception.3.An algorithm suppressing the temporal pumping artifact is proposed for the hierarchical B-pictures structure in H.264/AVC based on the content characteristics of the video.The regions with perceptible temporal pumping artifact are firstly obtained using the previously established algorithm.Secondly,spatial and temporal complexities of the regions with perceptible temporal pumping artifact are used for predicting the related quantization parameter(QP)threshold when the temporal pumping artifact just starts to be noticed.Then,the QP for each temporal layer is determined under the guidance of the predicted QP threshold.Experimental results show that compared with the currently used method of JVT-P014 QPC in the hierarchical B-pictures structure,the proposed method can obviously reduce the quality fluctuation of reconstructed videos,suppress the temporal pumping artifact and improve the subjective quality of reconstructed videos.4.Following the spirit of Lagrangian optimization,QPC methods which can effectively improve the rate-distortion performance of video coding are proposed for the random-access structure in H.265/HEVC.Firstly,the Lagrangian rate-distortion cost function is established for the hierarchical prediction structure considering the multiple temporal layers dependency relationships.Secondly,the established cost function is simplified using the characteristics obtained from the experimental observation.Thirdly,the simplified function is solved by using the Newton-Raphson method to determine QPs of different temporal layers.Experimental results show that compared with the currently used method of JCTVC-W1002 QPC in the random-access structure,the proposed QPC methods greatly improves the rate-distortion performance of video coding.5.An efficient frame level rate control algorithm for random-access structure in H.265/HEVC is proposed based on the previously proposed QPC methods.Under the guidance of the proposed QPC methods,the bit-rates can be allocated more reasonably for each temporal layer.And the temporal dependency is considered when calculating the Lagrange multiplier and the QP.Experimental results show that,compared with the algorithms in the HM,the proposed algorithm has more accurate rate control and higher coding rate-distortion performance.Although the temporal pumping artifact can be observed both in reconstructed videos using the proposed rate control algorithm and the algorithms in the HM,the perceived temporal pumping artifact degree of using the proposed algorithm is smaller.Especially for the head-and-shoulder video sequence,the videos to which the proposed algorithm is applied show better subjective quality.