Assessment of Image Quality using a Saliency Map

Abstract

Since a digital image is deteriorated by image processing such as compression and embedding of a digital watermark, image quality evaluation is required to indicate the degree of deterioration. PSNR and SSIM are widely used as existing objective image quality degradation evaluation methods. However, the results using PSNR and SSIM do not always agree with the subjective evaluation. In this study, we improve the existing physical evaluation and aim at the development of a physical evaluation method closer to the subjective evaluation. Then, since the part watched in the image is important in the subjective evaluation, a saliency map obtained using the features of the image is introduced into the image quality evaluation as perceptual information. The saliency map is a representation of a portion that is easy for a person to watch when recognizing an image. The purpose of this paper is to clarify how an evaluation index using a saliency map is related to conventional subjective evaluation and physical evaluation, by an image quality evaluation experiment.