小波域中 CSF 频率与方向加权的图像质量评价方法

摘要

为解决结构相似度算法的图像质量评价缺陷，提出一种在小波域中，根据 CSF 的频率与方向加权的图像质量评价方法。该方法将图像的小波域多分辨率特性与 CSF 的带通特性有机的相结合，得到一种基于结构相似度的评价新方法。算法首先将频率归一化的 CSF 曲线按照1倍频程分解成5份，然后进行5级 CDF 9/7双正交小波分解与重构，得到一个低频子带 LL 和多个中高频子带 HHi 、HLi 及 LHi ，根据不同的频率，不同的方向分配不同的权重，分别与相应的子频带 SSIM 值相乘，然后累加运算，得到小波域的加权图像结构相似度 WWSSIM 。实验表明，WWSSIM 方法的单调性、一致性、准确性评价指标较高，更接近人类主观视觉感受。%In order to solve the defects of SSIM algorithm for image quality evaluation ,frequency and direction weighed to CSF in Wavelet domain was proposed .This method organic combination the multi-resolution characteristics of wavelet with band-pass characteristic of CSF curve ,and get a new algorithm based on SSIM .Firstly ,the frequency normalized CSF curves according to 1 octave level into five parts ,and then 5 levels CDF 9/7 pairs of orthogonal wavelet decomposition and reconstruction ,to get a low-frequency sub-band LL and a plurality of medium or high frequency sub-band HHi ,HLi and LHi ,then assign different weights ac-cording to different frequencies ,different directions ,which were multiplied with the corresponding sub-band SSIM values ,and then do accumulate operations to obtain the weighted structural similarity in wavelet domain named WWSSIM .Experiments show that WWSSIM algorithm has high monotonicity ,consistency ,accuracy ,and closer to human subjective visual experience .