Publication: Enhancement of no-reference image quality assessment for contrast-distorted images using natural scene statistics features in Curvelet domain
No Thumbnail Available
Date
2017
Authors
Ahmed I.T.
Der C.S.
Journal Title
Journal ISSN
Volume Title
Publisher
Institute of Electrical and Electronics Engineers Inc.
Abstract
Contrast is a very important characteristic for visual perception of image quality. Some No-Reference Image Quality Assessment Algorithm NR-IQA metrics for Contrast-Distorted Images (CDI) have been proposed in the literature, e.g. Reduced-reference Image Quality Metric for Contrast-changed images (RIQMC) and NR-IQA for Contrast-Distorted Images (NR-IQACDI). Here, we intend to improve the assessment results of images available in databases such as TID2013 and CSIQ. Most of the NR-IQA metrics (e.g. NR-IQACDI) designed for CDI adopt features available in the spatial domain. This paper proposes to compliment it with feature in Curvelet domain which is powerful in capturing multiscale and multidirectional information in an image. We employed the Natural Scene Statistics (NSS) features in Curvelet domain originally recommended by Liu et al. (2014) which were found useful in the assessment of the quality of image distorted by compression, noise and blurring. Experiments were then conducted to assess the effect of incorporating these NSS features. The experimental results based on K-fold cross validation (K ranged from 2 to 10) and statistical test showed that the performance of NRIQACDI was improved. Future works include improvements of NRIQACDI, exploration of feature fusion methods and using a suitable feature selection method. � 2017 IEEE.
Description
Image enhancement; Systems engineering; Curvelets; Distorted images; Feature fusion method; Feature selection methods; Image quality assessment (IQA); K fold cross validations; Natural scene statistics; No-reference image quality assessments; Image quality