Sudhakar Nagalla & Ramesh Babu Inampudi International Journal of Image Processing (IJIP), Volume (8) : Issue (6) : 2014 468 Perceptual Weights Based On Local Energy For Image Quality Assessment Sudhakar Nagalla sudhakar.nagalla@becbapatla.ac.in Department of Computer Science and Engineering Bapatla Engineering College Bapatla, 522102, India Ramesh Babu Inampudi rinampudi@hotmail.com Department of Computer Science and Engineering Acharya Nagarjuna University Guntur, 522510, India Abstract This paper proposes an image quality metric that can effectively measure the quality of an image that correlates well with human judgment on the appearance of the image. The present work adds a new dimension to the structural approach based full-reference image quality assessment for gray scale images. The proposed method assigns more weight to the distortions present in the visual regions of interest of the reference (original) image than to the distortions present in the other regions of the image, referred to as perceptual weights. The perceptual features and their weights are computed based on the local energy modeling of the original image. The proposed model is validated using the image database provided by LIVE (Laboratory for Image & Video Engineering, The University of Texas at Austin) based on the evaluation metrics as suggested in the video quality experts group (VQEG) Phase I FR-TV test. Keywords: Image Quality, HVS, Full-reference Quality Assessment, Perceptual Weights. 1. INTRODUCTION Any image processing system should be aware of the impacts of processing on the visual quality of the resulting image. Numerous algorithms for image quality assessment (IQA) have been investigated and developed over the last several decades. The objective image quality measurement seeks to measure the quality of images algorithmically. Objective image quality metrics can be classified as full-reference in which the algorithm has access to the original (considered to be distortion free) image, no-reference in which the algorithm has access only to the distorted image and reduced-reference in which the algorithm has partial information regarding the original image. A comprehensive review of research and challenges in image quality assessment is presented in [1]. In [2], a number of simple statistical image quality metrics based on numerical errors are compared for gray scale image compression. These metrics include average difference, maximum difference, absolute error, mean square error (MSE), peak MSE, Laplacian MSE, histogram and Hosaka plot. It is observed that although some numerical measures correlate well with the human response for a specific compression technique, they are not found to be reliable for evaluation across various methods of compression. The most widely adopted statistical feature is the Mean Squared Error (MSE). However, MSE and its variants may not correlate well with subjective quality measures because human perception of image distortions and artifacts is unaccounted for. A detailed discussion on MSE is presented by Girod [3]. Most HVS based quality assessment metrics share an error-sensitivity based paradigm [4], which aims to quantify the strength of the errors between the reference and the distorted signals in a