Reduced reference image quality assessment based on statistics of edge

Abstract

ABSTRACT Objective Image Quality Assessment (IQA) model investigation is a hot topic in recent times. This paper proposed a novel and efficient universal Reduced Reference (RR) image quality assessment method based upon the statistics of edge discrimination. Firstly, binary edge maps created from the multi-scale wavelet transform modulus maxima were used as the low level feature to discriminate the difference between the reference and distorted image for IQA purpose. Then the gradient operator was applied on the binary map to produce the so called edge pattern map. The histogram of edge pattern map was used to verify the pattern of the edges of reference and distorted image, respectively. The RR features extracted from the histogram was used to discriminate the difference of edge pattern maps, and then form a new RR IQA model. Comparing to the typical RR model (Zhou Wang’s method, 2005), only 12 features (96 bits) are needed instead of 18 features (162 bits) in Zhou Wang et al.’s method with better overall performance. Keywords: wavelet, image quality assessment, modulus maxima, reduced reference 1. INTRODUCTION With the increasing use of applications based on visual da ta, objective image quality assessment has become an essential issue. Objective image quality models are mainly categorized into three types: Full reference (FR) models, for which the original image and the distorted image are required. Reduced reference (RR) models, for which a description of the original image into some parameters and the distorted imag e are both required. No reference (NR) models, which only require the distorted image. FR models have been investigated for a long time, and their performance has been validated across existed distortion types in realistic world. At present, several models have been used in the applications, such as Sarnoff JNDmetrix visual discrimination model (VDM) [1, 2] and Wang and Bovik’s SSIM [3]. Lately, Sheikh and Bovik proposed a competitive FR model named Visual Information Fide lity (VIF) [4,5]. It outperforms all state-of-the-art FR IQA models by a sizeable margin based on their report. However, in most cases, such as in-service visual quality monitoring, reference images cannot be achieved, and the application of FR IQA models is restricted greatly. It is im portant to develop more practical IQA models which need not use full reference image information. Conc erning to NR models, there is still no such efficient model that could deal with types of distortions. Even to specific distortion type, the performance of these NR models is not very satisfied [6,7,8]. RR models can be considered as the compromise between the unfeasibility of FR models and unavailability of NR models. It means that only partial information of image is demanded. The amount of this partial information is supposed to be much smaller than the reference image, and the distor ted image quality should be pr edicted with it. According to the report of video quality experts grou ps (VQEG), the investigation and validation of RR methods have been listed as one of the important issues for future directions [9, 10]. In general, for a typical RR system, features relate to the image perception quality is extracted at the sender side with limited bit rate. Then, these RR features are transmitted through an ancillary channel or are hidden in the reference