利用梯度奇异值分解的图像结构相似度评价

针对传统的图像质量评价方法中对图像结构信息的表征能力不足的问题,在研究了基于结构相似度和奇异值分解的两种图像评价方法的基础上,结合其不同特点提出了基于奇异值分解的结构相似度质量评价方法.该算法分别将参考图像和失真图像的梯度图像分成8×8大小的图像块,并对每一个图像块进行奇异值分解后计算对应图像块的奇异值相似性和各图像块局部方差分布的相似性,最后结合各图像块的奇异值相似性和图像的局部方差分布的相似来表征图像的畸变程度.对LIVE库中包括5种失真类型的982幅图像进行验证,其结果表明该评价方法能很好地对各种失真类型的图像进行评价,比峰值信噪比(PSNR)、结构相似度(SSIM)等算法的主客观一致性更好,更加符合人眼的视觉特性.

Similarity Assessment of Image Structural Quality with Gradient Singular Value Decomposition

Traditional methods of image quality assessment usually can not well express the structural information of the images,thus in this paper,firstly,two classical image quality assessment methods based on structural similarity and singular value decomposition were studied,and then an evaluation method based on the structural similarity of singular value decomposition was proposed by combining different features of the two mehtods.For this new method,the gradient images of the reference image and distortion image are divided into blocks with the size of 8×8,and then the similarities of both the Singular value and the local variance between reference image and the test image were calculated after singular value decomposition is performed on each block of the image.Finally,the degree of the image distortion was evaluated with the combination of the two methods.The experimental results on LIVE database which contains 5 different types of distortions and 982 distorted images show that the proposed method has a good evaluation of all kinds of distortion types and is highly consistent with human subjective evaluations.