Several Lifting Algorithms For Full-Reference Image Quality Assessment Models

In recent years,with the rapid development of multimedia technology,digital image processing has been widely used in various fields.Degradation of digital images is often inevitable due to image acquisition,transmission,and compression.In order to maintain or even improve the quality of the image,it is important to measure the quality of image.Therefore,it is very necessary to design a fast and efficient Image Quality Assessment(IQA)method.As the most commonly used IQA method,the full-reference was focused on this paper.Due to Structural Similarity(SSIM),Gradient Structural Similarity(GSSIM)can not well evaluate the quality of the severely distorted image and the high computational complexity of Feature Similarity(FSIM),we propose improved algorithms for feature extraction and feature pooling respectively.The main works of this paper are as follows:(1)A fast full-reference IQA algorithm is proposed,namely Gradient Weighted Lifting SSIM(GWL-SSIM)method.The algorithm is designed for the gray image or the luminance component of color image.Firstly,considering the gradient information can reflect the texture information of the image,and the nonlinearity of the image quality perception of the human visual system and the difference of the different component of the image,we extend the classic gradient to the generalized gradient,calculate the generalized gradient similarity.Then employ the generalized gradient similarity,contrast similarity and structural similarity to obtain a feature mapping image of local quality.Finally,considering the contribution of different regions to the visual perception of the image,the generalized gradient weighted pooling strategy is used to obtain the overall quality of the image.Numerical experiments,performed on six public databases and the different types of distortion,demonstrate that GWL-SSIM can achieve high computational efficiency and have a considerable evaluation result comparison with state-of-the-art IQA metrics.(2)A color IQA algorithm is proposed.On the basis of GWL-SSIM,using different color space transform and considering the chromaticity similarity of the image,we extend the GWL-SSIM to color image and obtain the color IQA based on different color space,namely GWL-SSIMc.Numerical experiments,performed on the different types of distortion of TID2008 database,demonstrate that for most distortion types,considering the chromaticity similarity of the color image can achieve the accuracy of IQA.(3)Two kinds of pooling strategies based on generalized mean are proposed to enhance the performance of SSIM,GSSIM and FSIM.Considering the impact of different pooling strategies on IQA,we promote the arithmetic average and harmonic average pooling strategy to the generalized mean and then obtain the generalized mean pooling strategy.Numerical experiments,performed on the TID2013 and TID2008 database,demonstrate that general mean pooling strategy can effectively improve the accuracy of IQA index,its evaluation results is more consistent with the human visual system.