No-Reference Stereoscopic Image Quality Assessment Based On Human Visual Characteristics

During image processing,various distortions are introduced inevitably into the original images in the process of creating,compressing,transmitting,reconstructing,rendering,and displaying.Therefore,the image quality assessment algorithm is urgent to effectively evaluate the visual quality of stereo image or video products.Therefore,it is important to establish an accurate and effective stereo image quality model in the digital multimedia era.A no-reference evaluation algorithm is proposed in this paper.The disparity,binocular fusion and saliency were taken account into and the disparity map,cyclopean map and saliency map were obtained successively by the corresponding algorithms.On the feature extraction method,the edge region and flat area were obtained by the phase congruency and the structure tensor.Then the texture features of sub-images were extracted by the gray-gradient co-occurrence matrix and the Tamura matrix.Finally,the training and testing of the model were conducted on two symmetrically distorted stereoscopic image databases.Experimental results illustrate that the prediction accuracy of the algorithm is fairly high.For asymmetrically distorted stereoscopic images,another no-reference evaluation algorithm is proposed based on singular value decomposition.First,considering the visual sensitivity to the spatial frequency variation and the binocular fusion,Gabor filtering was performed on the stereoscopic image,and a fusion strategy based on singular value decomposition was proposed to generate a cyclopean image of the left and right sub-image pair.Then,the proposed luminance-weighted histogram local binary pattern metric was used to extract the features of the cyclopean image,the left and right sub-images.The feature fusion and comparison were conducted on the two feature vectors corresponding to the left and right sub-images respectively.Especially,in order to measure the difference between the asymmetrical distorted sub-image pair,the image similarity metric was utilized to calculate the similarity between the left and right sub-image pair.The algorithm was tested on three stereoscopic image databases.The experimental results show that the proposed algorithm is of outstanding performance and superior to state-of-the-art IQA metrics.