压缩图像质量评价:一种带权的结构相似性方法

Compressed Image Quality Assessment:A Metric Based on Weighted Structural Similirity

图像质量评价是计算机视觉的一个重要领域,也是近年来的研究热点。在图像的传输与存储过程中,对压缩图像的客观质量评价在图像压缩系统中是必不可少的。均方误差与峰值信噪比虽然计算简便,但对于反映压缩图像的感知质量并不准确,与人眼视觉系统(HVS)也没有太好的一致性。近年来,该领域也出现了许多优秀的图像质量评价方法以求良好的表达图像质量与人眼感知之间的关系,如SSIM,IFC,VSNR等。这些算法性能卓越,但在压缩图像的质量评价方面仍有较大提升空间。图像在压缩过程中,高频部分由于DCT变换的特性会受到更大程度的失真,即图像不同部分的失真情况是不同的。从这个角度利用带权的结构相似性方法能良好的表达压缩图像的失真程度从而更好的反应人眼感知对图像质量的感知特性。该方法在LIVE,CSIQ,TID2008,TID2013上均比其他算法具有更好的性能。该方法也为压缩图像质量评价提供了一种新的思路,即从压缩算法本身会对图像信息的不同部分产生不同程度失真的特性出发来优化图像的质量评价,从而更好地反应HVS对图像的感知特性。

Image quality assessment(IQA) is an important area of computer vision in recent years. In the process of delivery and storage, objective quality assessment is essential for compressed images. Although the calculation of Mean Square Error(MSE)and Peak Signal to Noise Ratio(PSNR) are simple, their reflection to the perceptual quality of compressed images are not accurate. Recently, in order to improve the IQA ability, many excellent IQA methods are developed, such as SSIM, IFC, VSNR, etc.There are still a large ascension for compressed images although these metrics have good performances. As we known, the high frequency signals are more seriously distorted because of the DCT transform in image compression. Thus, based on the idea that the part of serious distortion can better present the degeneration of original images, we adopt the method as weighted structural similarity to evaluate the quality of compressed images. Extensive experiments have been performed on four benchmark databases, which demonstrate that the proposed method is more effective than a number of state-of-the-art IQA metrics.