基于视觉感知的梯度结构相似度图像质量评价*

尽管SSIM(Structural Similarity)图像质量评价算法结构简单，评价性能优于一般客观评价算法，但该算法没有考虑人类视觉系统HVS(human visual system)对视觉感知的影响，且其算法定义中对“结构信息”的表述过于简单，并不能完全描述自然图像的结构信息。在SSIM算法的基础上，结合亮度和对比度掩蔽等视觉感知信息构造视觉感知（Visual Perception）函数，提出基于视觉感知的梯度结构相似度评价方法VI_GSSIM（Visual Perception and Gradient based SSIM， VI_GSSIM）。该方法通过图像质量与图像内容和失真类型的相关性，结合图像的误差可视性与内容可视性构造视觉感知函数，对HVS底层视觉系统建模，同时利用梯度重新定义结构信息，得到基于视觉感知的梯度结构相似度模型，对图像进行质量评价。实验结果表明提出的VI_GSSIM算法比SSIM更符合人眼的视觉特性，尤其适合评价降质较严重的图像。

Image quality assessment based on visual perception and gradient structural similarity

Although the image quality evaluation algorithm based on structural information (SSIM) has a simple structure, and a better performance than other objective algorithms, but it does not take into account the effects on visual perception by human visual system, and the definition of the structure information of images is inaccurate and does not completely describe the structure of natural images. This paper proposed a new metric of image quality assessment based on visual perception and gradient structural similarity (VI_GSSIM), which is a combination of visual perception information based on luminance and contrast masking effect. This method uses the relevance between image quality and image content and distortion type of a degraded image, constructs visual perception with error visible and content visible to build a low-level HVS perception model, and then redefines structure information with image gradient to come into being the algorithm. The experimental results indicated that VI_GSSIM algorithm is more corresponding to the human visual system, especially suits to evaluate heavy distorting degraded image.