基于人眼失真敏感度的图像质量评价算法研究

人眼作为图像的最终观测者,对图像失真的敏感度会随着图像纹理的变化而变化。根据人眼视觉系统的这一特性,将图像的纹理信息引入到图像质量计算中,提出了一种基于人眼失真敏感度的结构相似度算法(DS-SSIM)。通过对LIVE DATABASE中5种失真类型共779张失真图像的仿真实验得出,相比PSNR,SSIM,ESSIM等算法,DS-SSIM算法的评价结果与人为主观评分具有更高的一致性,更加符合人眼视觉特性。

Image Quality Assessment Based on Distortion Sensitivity of Human Eyes

Human eyes, as the final observer of the image, are usually subjected to the images texture information to evaluate the distorted images quality. According to this feature of human visual system, the texture information is taken into the calculation of the image quality and a kind of structure similarity algorithm based on distortion sensitivity of human eyes (DS-SSIM) is put forward by us. Performance evaluation is performed on LIVE DATABASE of 779 distorted images including five kinds of distortion types. The results show that compared with PSNR, SSIM and ESSIM, DS-SSIM has a better consistency with human subjective ratings, more in line with the human visual system.