A Learning Based Contrast Specific no Reference Image Quality Assessment Algorithm

Contrast is one of the most important visual characteristics of an image that has a significant effect in understanding an image, however, due to different imaging conditions and poor devices, quality of image in terms of contrast will degrade. although, limited methods have been used to assess the quality of a contrast distorted images. Proper image contrast enhancement can increase the perceptual quality of most contrast distorted images. In this paper, assuming that the output images of a contrast enhancing algorithms have a quality such as a reference image, a learning-based contrast-specific no reference image quality assessment method is proposed. In the proposed method in this paper the image with the closest quality to the reference image is selected using a pre-trained classification network, and then the quality assessment is performed by comparing the enhanced image and the distorted image using structural similarity (SSIM) index. The functionality of the proposed method has been validated using three well-known contrast distorted image datasets (CSIQ, CCID2014 and TID2013).