No-reference quality assessment for contrast-distorted images based on multifaceted statistical representation of structure☆
Introduction
Contrast plays a significant role in the perception of image quality [1], [2]. Unfortunately, due to the imperfect capture device or inappropriate lighting conditions, contrast distortions usually occur in digital images. To address this problem, numerous contrast-enhancement algorithms have been proposed [3], [4]. Quality metrics for contrast-distorted images can be used as the benchmark, optimization and parameter-setting strategies for contrast-enhancement algorithms. Therefore, designing useful quality metrics for contrast-distorted images is of great significance.
The majority of existing quality metrics focus on images that are distorted by traditional distortions, such as blur [5], [6], JPEG compression [7], [8], and noise [9], [10], etc. There is a paucity of research on measuring contrast distortions. In general, image quality metrics can be classified into three categories, namely full-reference (FR), reduced-reference (RR), and no-reference (NR). For FR metrics, Wang et al. [1] proposed a Patch-based Contrast Quality Index (PCQI) metric by first representing image patches using signal strength, structure component, and mean intensity, and then severally calculating the average distances of these components between the reference and distorted images. Sun et al. [11] proposed to evaluate the quality of contrast-changed images (QCCI) using three perceptual aspects, i.e. contrast comparison, structure variation, and luminance change. For RR metrics, Gu et al. successively released two contrast-changed image databases, and accordingly proposed two Reduced-reference Image Quality Metrics for Contrast change (RIQMC) [13], [14]. The method [13] was presented based on multi-order statistics in salient regions, while the method [14] was proposed by further incorporating entropy features. For NR metrics, Fang et al. [2] proposed to first extract natural scene statistic features and then use the support vector regression to train a quality model. In [15], Gu et al. presented a NR Image Quality Metric for measuring Contrast distortions (NIQMC) from both local and global perspectives. The local contrast distortions were measured using the entropy of the regions with maximum information, while the global contrast distortions were measured by calculating the Jensen-Shannon divergence between the histograms of the contrast-distorted image and the uniformly distributed histogram of natural images.
In many contrast enhancement applications, high-quality reference images are not available, so NR metrics are needed. Although the aforementioned two NR metrics for contrast-distorted images have achieved notable progresses, the problem is still far from being solved. On the one hand, both metrics only calculate the distortions on gray-scale information, while ignore the effects of chromatic distortions on quality perception. Particularly, contrast distortions usually cause the change of chromatic information. Moreover, the chromatic information impacts the human perception of image quality [12]. It has been demonstrated that, in the first moment that human perceives natural scenes, 80% perceived information is chromatic information, and the percentage can be remained at about 50% in two minutes [12]. Moreover, Simone et al. [17] conducted a subjective experiment on both color and gray images, finding that the quality of the color version of a distorted image was commonly regarded to be worse than that of the corresponding gray-level image, except in the scenario that the test image was of extremely high or extremely low quality. This indicates that it is necessary to integrate the chromatic aspects in the quality model. On the other hand, existing NR quality metrics for contrast-distorted images only use the structure intensity for quality prediction, while ignore the fact that the human visual system (HVS) is also sensitive to the spatial distribution and orientation of structures [18], [19].
To address these issues, we propose a NR quality metric for contrast-distorted images by Multifaceted Statistical representation of Structure (MSS). To separate chromatic information from luminance, the input image is first transformed from RGB to S-CIELAB color space. Furthermore, the S-CIELAB color space has been demonstrated to be more consistent with the properties of the HVS [12]. With the consideration of the sensitivity of the HVS to structures [20], including spatial intensity, spatial distribution, and orientation [18], [19], the corresponding features that represent the three aspects of structural characteristics are extracted for structure representation. To map all features to an objective quality score, the back propagation (BP) neural network is utilized for training the quality model [27]. Experimental results on four contrast-distorted image databases demonstrate that the proposed method is superior to the relevant state-of-the-art quality metrics.
The remainder of this paper is organized as follows: Section 2 gives the details of the proposed method. Section 3 presents the experimental results and the corresponding analysis. Finally, Section 4 concludes the paper.
Section snippets
Proposed quality metric
The design philosophy of the proposed method is based on the following two facts. First, similar to luminance information, chromatic information also plays a significant role when human perceives natural scenes [12], and changes with the contrast distortions. Therefore, it is necessary to integrate the impacts of both luminance and chromatic aspects in the quality assessment of contrast distorted images. To this end, the input image is transformed from RGB to S-CIELAB color space to obtain the
Experimental settings
In this work, four public contrast-distorted image databases are used for performance test, namely CID2013 [14], CCID2014 [13], TID2013 [31], and CSIQ [32]. Particularly, the CID2013 and CCID2014 databases are two databases specifically designed for the quality assessment of contrast-changed images. The CID2013 database contains 400 contrast-changed images, which are generated from 15 original images by mean shifting or two transfer curves, while the CCID2014 database contains 655
Conclusion
With the considerations of the importance of chromatic information to visual perception and the sensitivity of the HVS to spatial distribution and orientation of structure, both of which are overlooked by existing quality metrics specific for contrast-distorted images, we propose a no-reference quality metric for contrast-distorted images based on the multifaceted statistical representation of structure. The statistical structural features are extracted from the three channels of S-CIELAB color
Conflict of interest
No conflict of interest exits in the submission of this manuscript.
Acknowledgment
This work was supported by the National Natural Science Foundation of China under Grants 61771473 and 61379143, Natural Science Foundation of Jiangsu Province under Grant BK20181354, Six Talent Peaks High-level Talents in Jiangsu Province under Grant XYDXX-063, and the Qing Lan Project of Jiangsu Province.
References (54)
- et al.
Image contrast enhancement based on intensity expansion-conpression
J. Vis. Commun. Image Represent.
(2017) - et al.
Contrast enhancement of noisy low-light images based on structure-texture-noise decomposition
J. Vis. Commun. Image Represent.
(2017) - et al.
Blind image sharpness assessment based on local contrast map statistics
J. Vis. Commun. Image Represent.
(2018) - et al.
GridSAR: grid strength and regularity for robust evaluation of blocking artifacts in JPEG images
J. Vis. Commun. Image Represent.
(2015) - et al.
Orientation selectivity based visual pattern for reduced-reference image quality assessment
Inform. Sci.
(2016) - et al.
Detecting seam carving based image resizing using local binary patterns
Comput. Sec.
(2015) - et al.
Blind image quality assessment by relative gradient statistics and adaboosting neural network
Signal Proces.: Image Commun.
(2016) - et al.
Boosted decision trees as an alternative to artificial neural networks for particle identification
Nucl. Instrum. Methods Phys. Res. A Accel. Spectrom. Detect. Assoc. Equip.
(2005) - et al.
Image database TID2013: peculiarities, results and perspectives
Signal Process. Image Commun.
(2015) - et al.
An efficient and effective blind camera image quality metric via modeling quaternion wavelet coefficients
J. Vis. Commun. Image Represent.
(2017)
A patch-structure representation method for quality assessment of contrast changed images
IEEE Signal Process. Lett.
No-reference quality assessment of contrast-distorted images based on natural scene statistics
IEEE Signal Process. Lett.
A fast approach for no-reference image sharpness assessment based on maximum local variation
IEEE Signal Process. Lett.
Learning structural regularity for evaluating blocking artifacts in JPEG images
IEEE Signal Process. Lett.
No-reference quality assessment of noise-distorted images based on frequency mapping
IEEE Access
The relative impact of ghosting and noise on the perceived quality of MR images
IEEE Trans. Image Process.
Full-reference quality assessment of contrast changed images based on local linear model
Image quality assessment based on S-CIELAB model
Signal Image Video Process.
The analysis of image contrast: from quality assessment to automatic enhancement
IEEE Trans. Cybern.
Subjective and objective quality assessment for images with contrast change
No-reference quality metric of contrast-distorted images based on information maximization
IEEE Trans. Cybern.
A subjective study of the influence of color information in visual quality assessment of high resolution pictures
When spatial distribution unites with spatial contrast: an effective blind image quality assessment model
IET Image Process.
Image quality assessment based on local orientation distribution
Image quality assessment using directional anisotropy structure measurement
IEEE Trans. Process.
Orientation-selective adaptation to first- and second-order patterns in human visual cortex
J. Neurophysiol.
Image quality assessment based on gradient similarity
IEEE Trans. Image Process.
Cited by (5)
No-reference quality metric for contrast-distorted image based on gradient domain and HSV space
2020, Journal of Visual Communication and Image RepresentationCitation Excerpt :To address the issue of contrast distortion, numerous contrast-enhancement algorithms have been proposed [20,21]. And quality metrics for contrast-distorted images provide the evaluation criteria and optimising direction for contrast-enhancement algorithms [22]. However, there are few algorithms for contrast distortion.
Online Low-Light Sand-Dust Video Enhancement Using Adaptive Dynamic Brightness Correction and a Rolling Guidance Filter
2024, IEEE Transactions on MultimediaPerceptual Quality Assessment of Enhanced Colonoscopy Images: A Benchmark Dataset and an Objective Method
2023, IEEE Transactions on Circuits and Systems for Video TechnologyNO-REFERENCE QUALITY ASSESSMENT OF MEDICAL IMAGES USING CONTRAST ENHANCEMENT
2023, Journal of Theoretical and Applied Information TechnologyNo reference quality assessment method for contrast-distorted images based on three elements of color
2022, Beijing Hangkong Hangtian Daxue Xuebao/Journal of Beijing University of Aeronautics and Astronautics
- ☆
This paper has been recommended for acceptance by Xuanqin Mou.