液晶与显示, 2018, 33 (10): 877, 网络出版: 2018-12-18
基于稀疏表示的SOM多失真图像质量评价方法
Multi-distorted image quality assessment algorithm based on sparse representation and SOM
图像质量评价 多失真图像 稀疏表示 聚类 交叉验证 image quality assessment multi-distorted images sparse representation clustering cross-validation
摘要
针对非线性回归下客观评分与主观评分一致性差的问题,本文提出一种基于稀疏表示的SOM多失真图像质量评价方法。首先,将参考图像及失真图像应用独立变量分析进行稀疏化表示,应用稀疏表示下的参考图像与失真图像间的结构相似度描述失真图像的质量,再使用SOM聚类算法和交叉验证方法提高非线性回归下的客观评分与主观评分之间的一致性。最后,在LIVE2, TID2013及IVC数据库中的实验结果显示,所提评价模型性能优越;3种数据库的平均结果说明,该文方法的总体性能高于现有的经典算法,表明该文方法能够很好地反映图像的视觉感知效果。通过对比时间效率,该方法基本能够满足实际要求,具有较高的适用性。
Abstract
Due to the problem of poor consistency between objective score in nonlinear regression and subjective scores, a multi-distorted image quality assessment method based on sparse representation and SOM is presented in this paper. Firstly, the reference image and the multi-distorted image are represented sparsely by independent component analysis, and the structural similarity index between the reference image and the distorted image under the sparse representation is computed to describe the quality of multi-distorted image. Secondly, the consistency between objective score in nonlinear regression and subjective score is improved by the SOM and cross-validation algorithms. Finally, the experimental results in LIVE2, TID2013 and IVC databases show that the proposed evaluation model has good performance. The average result of 3 kinds of databases shows that the overall performance of the method is higher than the existing classical algorithm, which indicates that the proposed method can reflect the visual perception of the image well. Comparison with the time efficiency, the proposed algorithm can basically meet the practical demand and has high practicability.
王春哲, 安军社, 姜秀杰, 李杰, 张羽丰. 基于稀疏表示的SOM多失真图像质量评价方法[J]. 液晶与显示, 2018, 33(10): 877. WANG Chun-zhe, AN Jun-she, JIANG Xiu-jie, LI Jie, ZHANG Yu-feng. Multi-distorted image quality assessment algorithm based on sparse representation and SOM[J]. Chinese Journal of Liquid Crystals and Displays, 2018, 33(10): 877.