Please use this identifier to cite or link to this item: http://localhost/handle/Hannan/642965
Title: Analysis of Distortion Distribution for Pooling in Image Quality Prediction
Authors: Ke Gu;Shiqi Wang;Guangtao Zhai;Weisi Lin;Xiaokang Yang;Wenjun Zhang
subject: distortion distribution|ranking-based weighting (RW)|entropy gain multiplier (EGM)|frequency variation-induced adjuster (FVA)|multi-scale (MS)|Image quality assessment (IQA)|pooling
Year: 2016
Publisher: IEEE
Abstract: Image quality assessment (IQA) has been an active research area during last decades. Many existing objective IQA models share a similar two-step structure with measuring local distortion before pooling. Compared with the rapid development for local distortion measurement, seldom effort has been made dedicated to effective pooling schemes. In this paper, we design a new pooling model via the analysis of distortion distribution affected by image content and distortion. That is, distributions of distortion position, distortion intensity, frequency changes, and histogram changes are comprehensively considered to infer an overall quality score. Experimental results conducted on four large-scale image quality databases (LIVE, TID2008, CSIQ, and CCID2014) concluded with three valuable findings. First, the proposed technique leads to consistent improvement in the IQA performance for studied local distortion measures. Second, relative to the traditional pooling, the performance gain of our algorithm is beyond 15% on average. Third, the best overall performance made by the proposed strategy outperforms state-of-the-art competitors.
URI: http://localhost/handle/Hannan/176171
http://localhost/handle/Hannan/642965
ISSN: 0018-9316
1557-9611
volume: 62
issue: 2
Appears in Collections:2016

Files in This Item:
File Description SizeFormat 
7390011.pdf1.88 MBAdobe PDFThumbnail
Preview File
Title: Analysis of Distortion Distribution for Pooling in Image Quality Prediction
Authors: Ke Gu;Shiqi Wang;Guangtao Zhai;Weisi Lin;Xiaokang Yang;Wenjun Zhang
subject: distortion distribution|ranking-based weighting (RW)|entropy gain multiplier (EGM)|frequency variation-induced adjuster (FVA)|multi-scale (MS)|Image quality assessment (IQA)|pooling
Year: 2016
Publisher: IEEE
Abstract: Image quality assessment (IQA) has been an active research area during last decades. Many existing objective IQA models share a similar two-step structure with measuring local distortion before pooling. Compared with the rapid development for local distortion measurement, seldom effort has been made dedicated to effective pooling schemes. In this paper, we design a new pooling model via the analysis of distortion distribution affected by image content and distortion. That is, distributions of distortion position, distortion intensity, frequency changes, and histogram changes are comprehensively considered to infer an overall quality score. Experimental results conducted on four large-scale image quality databases (LIVE, TID2008, CSIQ, and CCID2014) concluded with three valuable findings. First, the proposed technique leads to consistent improvement in the IQA performance for studied local distortion measures. Second, relative to the traditional pooling, the performance gain of our algorithm is beyond 15% on average. Third, the best overall performance made by the proposed strategy outperforms state-of-the-art competitors.
URI: http://localhost/handle/Hannan/176171
http://localhost/handle/Hannan/642965
ISSN: 0018-9316
1557-9611
volume: 62
issue: 2
Appears in Collections:2016

Files in This Item:
File Description SizeFormat 
7390011.pdf1.88 MBAdobe PDFThumbnail
Preview File
Title: Analysis of Distortion Distribution for Pooling in Image Quality Prediction
Authors: Ke Gu;Shiqi Wang;Guangtao Zhai;Weisi Lin;Xiaokang Yang;Wenjun Zhang
subject: distortion distribution|ranking-based weighting (RW)|entropy gain multiplier (EGM)|frequency variation-induced adjuster (FVA)|multi-scale (MS)|Image quality assessment (IQA)|pooling
Year: 2016
Publisher: IEEE
Abstract: Image quality assessment (IQA) has been an active research area during last decades. Many existing objective IQA models share a similar two-step structure with measuring local distortion before pooling. Compared with the rapid development for local distortion measurement, seldom effort has been made dedicated to effective pooling schemes. In this paper, we design a new pooling model via the analysis of distortion distribution affected by image content and distortion. That is, distributions of distortion position, distortion intensity, frequency changes, and histogram changes are comprehensively considered to infer an overall quality score. Experimental results conducted on four large-scale image quality databases (LIVE, TID2008, CSIQ, and CCID2014) concluded with three valuable findings. First, the proposed technique leads to consistent improvement in the IQA performance for studied local distortion measures. Second, relative to the traditional pooling, the performance gain of our algorithm is beyond 15% on average. Third, the best overall performance made by the proposed strategy outperforms state-of-the-art competitors.
URI: http://localhost/handle/Hannan/176171
http://localhost/handle/Hannan/642965
ISSN: 0018-9316
1557-9611
volume: 62
issue: 2
Appears in Collections:2016

Files in This Item:
File Description SizeFormat 
7390011.pdf1.88 MBAdobe PDFThumbnail
Preview File