基于视觉感知的全参考图像质量评价算法

Full-reference image quality assessment algorithm based on visual perception

针对传统图像质量评价算法不能有效反映图像的视觉感知质量,存在主观一致性较差的问题,提出了一种基于视觉感知的全参考图像质量评价算法。利用人眼对于图像边缘极为敏感的视觉系统感知特性,构造区域掩膜按感知重要性对图像进行分割,经方向性对比度敏感函数滤波后加权,构造出一个与人眼主观感受相一致的客观图像质量评价指标。基于CSIQ数据库进行了大量的实验并与其他算法进行比较,实验结果表明,本文算法对于多种类型失真图像的评价结果均具有较高的主观一致性,能准确反映人眼的视觉感知质量,尤其是针对JPEG2000压缩失真,性能大幅优于其他算法。

Traditional quality assessment algorithms cannot effectively reflect the image quality of visual perception and perform poor consistency. To solve this problem,a novel image quality assessment algorithm based on visual perception is proposed in this paper. According to the characteristic that the eye is extremely sensitive to image edges,the image segmentation and the directional contrast sensitivity function are used,and then the signal-tonoise ratio of reference image and the distortion image are weighted,finally,an image quality index with high consistency is gotten. The experiments results on database CSIQ show that the algorithm proposed in this paper performs higher subjective consistency in various types of image distortion than other algorithms,especially in JPEG2000 compression distortion.