Image Quality Assessment Based on Perceptual Structural Similarity D. Venkata Rao and L. Pratap Reddy Bapatla Engineering College, Bapatla, India JNTU College of Engineering, Hyderabad, India dv2002in@yahoo.co.in,pratplr@rediffmail.com Abstract. We present a full reference objective image quality assess- ment technique which is based on the properties of the human visual system (HVS). It consists of two major components: 1) structural simi- larity measurement (SSIM) between the reference and distorted images, mimicking the overall functionality of HVS in a top down frame work. 2) A visual attention model which indicates perceptually important regions in the reference image based on the characteristics of intermediate and higher visual processes through the use of Importance Maps. Structural similarity in a region is weighted, depending on the perceptual impor- tance of the region to arrive at Perceptual Structural Similarity Metric (PSSIM) indicative of the image quality. Keywords: Objective image quality, HVS, structural distortion, percep- tually important regions. 1 Introduction The role of images in present day communication has been steadily increasing. In this context the quality of an image plays a very important role. Different stages and multiple design choices at each stage exist in any image processing system. They have direct bearing on the quality of the resulting image. Unless we have a quantitative measure for the quality of an image, it becomes difficult to design an ideal image processing system. Though subjective quality assessment is an alternative, it is not feasible to be incorporated into real world systems. Hence, objective quality metrics play an important role in image quality assessment. In the last two decades a lot of objective metrics have been proposed [1-7] to assess image quality. The most widely adopted statistics feature is the Mean Squared Error (MSE). However, MSE and its variants do not correlate well with subjective quality measures because human perception of image distortions and artifacts is unaccounted for. MSE also not good because the residual image is not uncorrelated additive noise,it also contains components of the original image. A detailed discussion on MSE is given by Girod [8]. A major emphasis in recent research has been given to a deeper analysis of the Human Visual System (HVS) features [1]. There are lot of HVS characteristics [9] that may influence the human visual perception on image quality. Although A. Ghosh, R.K. De, and S.K. Pal (Eds.): PReMI 2007, LNCS 4815, pp. 87–94, 2007. c  Springer-Verlag Berlin Heidelberg 2007