A Novel Full Reference Metric for Image Quality Assessment Based on Human Vision System

Abstract

With the tremendous growth in field of digital image processing, there is a need to store and maintain digital images efficently. Digital images are captured, processed, compressed, stored and transmitted through various devices. During these stages of processing, there is a need to maintain quality of the images because various distortions may take place and affect the quality of the image. To assess the quality of an image there are subjective and objective methods available. Subjective methods are generally time consuming and difficult to use in autonomous systems. Objective image quality assessment methods are more used these days to evaluate the quality of the image. These objective image quality assessment methods are based on the situation that whether the reference image is available, partially available or just not available. Today, most of evaluations of images are based on the assumption that reference image is available this is called full reference method. There are many metrics based on full reference approach to determine the quality of an image. Some metrics are mathematical like Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR) etc. These metrics are not able to correlate well with human perception. Some metrics are based on Human Vision System (HVS) like Structural Similarity Index (SSIM), Visual Image Fidelity (VIF), Visual Discrimination Model (VDM), Perceptual Difference Method (PDiff) etc. These metrics also have some limitations in evaluating the quality of the image. There are many full reference metrics implemented for assessing the quality of an image based on high level and low features of an image based on HVS. These are implemented by different metrics. In this thesis effort is made to implement a new full reference metric by integrating the human vision based metrics including SSIM, VIF, VDM and Pdiff. Further the quality of image has been tested on available Laboratory for Image & Video Engineering (LIVE) Image Quality Assessment Database. The quality index of the image has been observed and reported in this thesis.