基于变差函数全局纹理增强的结构相似度图像质量评价

为解决结构相似度算法的图像质量评价缺陷,提出了一种基于变差函数全局纹理增强的结构相似度图像质量评价。该方法首先利用改进的对数变差函数模型提取原图像和失真图像在水平0°、垂直90°和对角45°、135°四个方向的纹理信息特征,然后分别求出对应的纹理增强图像,最后改进SSIM中的结构信息来确定纹理区域的明显失真,计算得到整幅图像的VSSIM值。目前大多数的全参考评价方法不能对数据库中的所有失真类型进行评价,只能对某一类固定的失真类型来评价。本方法对LIVE数据库中的五种失真类型都适用,仿真实验表明,该算法对五种不同失真类型的评价结果具有一定的合理性,并且与主观评价数据库较为一致,其性能也优于其他质量评价模型。

Image quality assessment based on structure similarity with variation function global texture enhancement

To solve the defects of image quality assessment by structure similarity algorithms, we propose an image quality assessment method based on structure similarity with variation function texture enhancement. Firstly the improved logarithmic variation function model is utilized to extract the texture information characteristics of the original image and the distorted images in the four directions of horizontal 0°, vertical 90°, diagonal 45° and diagonal 135°, and the corresponding texture enhancement images are calculated respectively. Then the structure information of the improved SSIM is used to determine the obvious distorted texture region, and the VSSIM value of the whole image is calculated. At present most of the existing evaluation methods are mainly used for a particular type of distortion , while the proposed method can be applied to all the five types of distortion in LIVE database. Experimental results show that the distortion evaluation results are stable, reasonable and relatively consistent with subjective evaluation database, and its performance is superior to other quality evaluation models.