深度视觉特征与策略互补融合的图像质量评价

Image Quality Assessment Based on Complementary Pooling of Deeply Visual Feature and Strategy

视觉多特征融合方法未考虑图像不同特征之间和不同评价算法之间的视觉互补性.通过融合人类视觉系统前端生理感知和后端心理处理特性,文中提出深度视觉特征互补融合(CPDVF)的图像质量评价方法.CPDVF深度提取图像的多通道直方图统计和多通道梯度结构这2种互补视觉特征,并进行深度视觉处理.然后设计局部失真度评价和局部相似度评价2种互补算法,分别对失真图像的上述互补视觉特征进行评价.最后联合视觉心理特性和回归函数,融合2种特征评价,获得失真图像质量的客观评价.实验表明,相比特征相似度、视觉显著等多特征联合方法,文中方法在准确度、单调性和可靠性指标上优势明显.

In methods of visual multiple feature pooling, visual complementary between different image features and different assessment algorithms is not taken into account. A method for complementary pooling of deeply visual feature（CPDVF） is proposed based on the integration of physiological perception of front human visual system（HVS） and psychological processing of back HVS in this paper. Firstly, two kinds of complementary features, the histogram statistics and gradient structure for visual multi-channel, are extracted and deeply processed based on visual characteristics. Secondly, the local distortion algorithm for visual histogram pooled contrast（VHPC） assessment and the local complementary similarity algorithm for visual gradient pooled contrast（VGPC） assessment are proposed. Finally, the distorted image qualityis obtained with pooling of VHPC and VGPC based on psychological characteristics and regression function. The experimental results show that the CPDVF is superior to feature similarity and visual saliency in accuracy, stability and monotonicity.