SUBJECTIVE AND OBJECTIVE QUALITY ASSESSMENT FOR IMAGES WITH CONTRAST CHANGE Ke Gu, Guangtao Zhai, Xiaokang Yang, Wenjun Zhang, and Min Liu Institute of Image Communication and Information Processing, Shanghai Jiao Tong University, Shanghai, China Shanghai Key Laboratory of Digital Media Processing and Transmissions ABSTRACT It is widely known that, for most natural images, appropriate contrast enhancement can usually lead to improved subjec- tive quality. Despite of its importance to image processing, contrast change has largely been overlooked in the current research of image quality assessment (IQA). To fill this void, in this paper we first report a new and dedicated contrast- changed image database (CID2013). The CID2013 database is composed of four hundred contrast-changed images of fifteen original natural images and the mean opinion scores (MOSs) recorded from twenty-two inexperienced viewers. We then proposed a novel reduced-reference image quality metric for contrast-changed images (RIQMC) using entropies and order statistics of the image histograms. Experimen- tal results on the CID2013, TID2008, and CSIQ databases demonstrate that the proposed RIQMC metric outperforms some mainstream image quality assessment methods. Index Terms— Contrast-changed image database, mean opinion score (MOS), image quality assessment (IQA), en- tropy, order statistics 1. INTRODUCTION The research of image quality assessment (IQA) aims to de- velop an image quality metric that is well correlated with the human subjective scores. Limited by the dependence on sub- jective quality databases, most of existing IQA research fo- cuses on compression artifacts, noise injection and blurring. Although contrast enhancement is an active topic in image processing [1]-[4], contrast as an important feature of image, has been largely overlooked in the currently research of IQA. And this may be partially because there lacks of a dedicated and large IQA database of images with contrast-change. We introduce in this paper a new contrast-changed im- age database (CID2013) 1 to facilitate the research of contrast related IQA. The CID2013 database includes four hundred contrast-changed versions of fifteen natural images chosen from the Kodak database [5]. Twenty-two inexperienced ob- servers participated the subjective viewing tests designed ac- 1 CID2013 database will be open to the public soon. cording to ITU-R BT.500-12 [6] and the mean opinion scores (MOSs) are recorded. Since the subjective quality test is always laborious and impractical for real-world image processing systems, we fur- ther proposed a reduced-reference image quality metric for contrast-changed images (RIQMC) using entropies and order statistics of the image histograms. Entropy [7] measures the average unpredictability of random variables and so we use image entropy as an indicator of image contents. Order statistics on histograms have widely been used in many contrast related image processing tasks. Obviously, the mean, or the first order statistic of an image, determines the overall brightness of the image histogram [1]. In the re- cently proposed optimal contrast-tone mapping (OCTM) al- gorithm [4], the concept of expected context-free contrast is defined as a function of variance (second order statistic) of the histogram. A surface perception model [8] suggests that there exists connection between human perception of surface glossiness and skewness (third order statistic). Some recent work on natural image analysis reveals that kurtosis (fourth order statistic) captures some intrinsic properties of natural images [9]. As inspired by the entropy and order statistics based methods, we design the RIQMC algorithm through lin- early combining the entropy and the first four order statistics of the image histogram. The RIQMC method is compared with some well-known benchmark full-reference IQA meth- ods such as PSNR, SSIM [10], MS-SSIM [11], IW-SSIM [12], and MAD [13]. The SSIM algorithm and its classical improved versions are considered quite relevant here because Fig. 1. The fifteen lossless natural color images in Kodak image database [5].