1. No-reference image quality assessment with visual pattern degradation.
- Author
-
Wu, Jinjian, Zhang, Man, Li, Leida, Dong, Weisheng, Shi, Guangming, and Lin, Weisi
- Subjects
- *
IMAGE quality analysis , *VISUAL perception , *IMAGE processing , *DOCUMENT clustering , *VISUAL cortex , *DATA mining , *PIXELS - Abstract
The technology of image quality assessment (IQA) is urgently demanded in perceptual-orientated image processing systems. Due to the lack of guidance from the reference information, it is challenging for the no-reference (NR) IQA to perform consistently with the human perception. It is well known that even though there exists no reference information, the human visual system can still effectively predict the image quality under the guidance of the prior knowledge. Thus, we try to build a visual content based prior knowledge database to guide the NR IQA. Cognitive researches state that the primary visual cortex presents substantially orientation selectivity, within which the structural information is extracted for visual perception. Inspired by this, the correlations among pixels in a local region are firstly analyzed for structure extraction. By mimicking the excitatory/inhibitory interactions among neurons, the relationships among pixels are estimated as their orientation similarities. Next, by aligning all of the correlations in a local region, a novel orientation similarity based pattern (OSP) is designed. Moreover, by merging the similar patterns with the clustering procedure, a set of fundamental OSPs are acquired. And then, the visual structure degradations are analyzed on these fundamental OSPs, and a prior knowledge database is learned. With the guidance of the knowledge database, a new visual pattern degradation (VPD) based NR-IQA model (NRVPD) is built. Finally, the performance of the proposed NRVPD model is verified on large benchmark IQA databases, and the experimental results demonstrate that the NRVPD performs highly consistent with the subjective perception. [ABSTRACT FROM AUTHOR]
- Published
- 2019
- Full Text
- View/download PDF