Title: New image quality assessment metric based on distortion classification

Authors: Xin Jin; Mei Yu; Shanshan Liu; Yang Song; Gangyi Jiang

Addresses: Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China ' Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China; National Key Lab of Software New Technology, Nanjing University, Nanjing 210093, China ' Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China ' Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China ' Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China; National Key Lab of Software New Technology, Nanjing University, Nanjing 210093, China

Abstract: Image quality assessment (IQA) has been a crucial task in image proceeding applications. In this paper, we propose an adjusted structure similarity (ASSIM) metric by considering the properties of different distortion types. In detail, we firstly define four particular features in accordance with different characteristics of distortion types. Secondly, a support vector machine-based multi-classifier is built to identify the distortion type of each distorted image on the basis of those four features. Thirdly, the well-known SSIM metric is adjusted by reallocating weighting values among its three evaluating factors for each distortion types. Finally, by combining the distortion classification (DC) and ASSIM, the subjective quality of each image is predicted. The experimental results derived from public test image databases show that the proposed DC method outperforms existing method across databases and degradations. Moreover, the proposed ASSIM metric can achieve high consistency with subjective perception.

Keywords: image processing; image quality assessment; IQA; distortion classification; support vector machine; SVM; multi-classifier.

DOI: 10.1504/IJICT.2017.086251

International Journal of Information and Communication Technology, 2017 Vol.11 No.2, pp.243 - 259

Received: 25 Oct 2014
Accepted: 21 Nov 2014

Published online: 11 Aug 2017 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article