No-reference stereoscopic image quality assessment based on contourlet transform
Chen Li, Shuiyuan Yu, Zhiguo Hong · 2016
According to the visual characteristics of the human eye perception, a no-reference (NR) stereoscopic image quality assessment (SIQA) method is proposed based on contourlet transform, which considered from the quality of the viewpoint images and the depth perception. Firstly, the subband energy of the left and right viewpoint images is extracted as viewpoints feature by contourlet decomposition. Then, by computing the parallax of the stereoscopic image, we can work out the matching view area, from which the subband energy is calculated as parallax feature by contourlet decomposition. At last, we use support vector regression (SVR) to train the relationship model between the perceptual features of stereoscopic image and subjective scores, and the learnt model is utilized to predict the quality of test stereoscopic images. Experimental results show that the Pearson Linear Correlation Coefficient (PLCC) and Spearman Rank Order Correlation Coefficient (SROCC) of the proposed method are higher than 0.9 in LIVE 3D image quality database, which indicates that compared with other methods, our method is better consistent with subjective assessment of stereoscopic images.