Research On Comfort Factors And Quality Evaluation Methods Of Stereo Images

The degradation of stereo images,the lack of stereo sense and the lack of viewing comfort have become important constraints for the promotion and popularization of stereo products.Therefore,how to effectively evaluate the comfort of stereo images and stereo image quality has become an important research direction in related fields at home and abroad.The main work of the paper is as follows.Firstly,based on visual saliency,a quantitative study of comfort chroma/comfort saturation of stereo images based on salient region is proposed.Firstly,the method combines stereo disparity map with plane salient map to get stereo salient map,and then optimizes it by fuzzy membership and mask to get the final salient stereo image.Then,we use the coarse to fine step by step approximation method to obtain experimental data for subjective experiments.Finally,the comfortable chroma/saturation matching graph and difference graph of different scene salient stereo images are obtained by analyzing and processing the experimental data.Second,a dual channel stereo image quality assessment method based on sparse dictionary learning is proposed.One channel combines the visual attention mechanism to get the initial salient map,and then optimizes it with the characteristics of center bias and concave center to get the final salient map.Then,the salient map is trained by sparse dictionary to obtain salient dictionary.The other channel transforms the reference stereo image pairs into SIFT features,and then improves them.The SIFT dictionary is trained by sparse dictionary training.In the test phase,the sparse coefficients of the reference image and the distorted image are obtained by sparse coding with the trained dictionary,and the similarity index of the sparse coefficients is defined to measure the information difference between the reference image and the distorted image.Thirdly,the thesis simulates the mechanism of human brain processing stereo images,designs a stereo image fusion algorithm,and then proposes two stereo image quality evaluation methods based on this fusion image.One of the methods(C-SIQA)uses a sparse dictionary to reconstruct the fused image sparsely.In order to compensate for the loss of information in the process of reconstruction,the corresponding color fusion image is proposed to compensate the information beforefeature extraction.Then spatial entropy and spectral entropy are used to extract the features of the reconstructed fusion image and the corresponding fusion image.Finally,the extracted spatial entropy and spectral quotient features are weighted and sent to SVR to obtain the objective mass fraction of the stereo image.The structure of CS-SIQA is basically the same as that of C-SIQA,but the difference is that it proposes to use the salient right view of stereo image to replace the corresponding fusion image for information compensation.In order to verify the reliability and validity of the objective evaluation method for stereo image quality,experiments were carried out on two open LIVE databases.The experimental results show that the evaluation results of the above stereo image quality evaluation methods have a good correlation with the subjective scores.Compared with some existing stereo image quality evaluation methods,they are more consistent with human visual perception,especially for the asymmetric distortion data.