Stereoscopic image quality assessment based on visual threshold and channel fusion

According to the different sensitivities of human eyes for different color channels of color images and the effect of tolerable perception range of human eyes on color image perception, a color image objective quality assessment method based on visual threshold and channel integration is proposed. Visual threshold characteristics of human eyes are used to determine whether the distortion of a stereoscopic image can be perceived. If the degree of distortion is within the tolerable range of human eyes, the distortion is ignored. Singular values of difference maps between original and distorted images with respect to the left and right views are compared with singular values of the visual threshold map of original images, so as to assess the quality of the left and right views, respectively. For original and distorted stereoscopic images, the difference between their left view and the right view are calculated, and the difference map of the above two differences is further calculated. The singular values of the difference map are then compared with singular value of binocular just noticeable difference to assess the quality of stereo perception. For different types of distortion, the weights for fusing the assessments of the left and right views, and the weights for fusing the assessments of the left-right views and stereo perception are different. The experimental results on 312 stereoscopic images distorted with Gaussian blur, Gaussian white noise, JPEG, JP2K and H.264 show that the proposed objective model can achieve more than 0.94 Pearson Linear Correlation Coefficient(CC) and general Spearman Rank Order Correlation Coefficient(SROCC), respectively, and the overall Rooted Mean Square Error(RMSE) is less than 5.9, which means that the assessment score obtained by the proposed model is well consistent with that obtained by human subjective perception.