基于像素失真耦合边缘特征融合的无参考质量评价

为了适应多种类型的模糊图像进行质量评价,提高评价模型对图像模糊和振铃的洞察能力,提出了一种像素失真与边缘特征融合的无参考质量评价算法.首先,根据像素失真理论,计算图像像素的标准差和绝对差分值,提取图像的像素特征;然后,计算图像水平和垂直方向的过零率,并利用边缘保持滤波器对图像边缘信息进行测量,精确提取图像的边缘特征;最后,利用提取的像素特征和边缘特征,定义特征融合函数,并引入粒子群优化(PSO)对融合函数参数进行优化,提高对图像模糊和振铃的洞察能力,根据融合特征构建图像质量评价模型.与当前无参考质量评价算法比较,所提算法能够有效地对JPEG(Joint Photographic Experts Group)、JPEG2000(Joint Photographic Experts Group 2000)、模糊等失真图像进行质量评价,评价指标CC(Correlation Coefficient)与SROCC(Spearman Rank-order Correlation Coefficient)达0.9477和0.9153.该算法与主观评价方法具有较好的一致性,能够较好地适用于多种类型的失真图像评价.

No-reference quality evaluation based on pixel distortion and edge feature fusion

In order to evaluate the quality of various types of fuzzy images and improve the ability of insight into the evaluation model of the image blurring and ringing, a no-reference quality evaluation algorithm based on the fusion of pixel distortion and edge feature is proposed. Firstly,according to the theory of pixel distortion,standard deviation and absolute difference value of image pixels is calculated,and pixel feature is extracted. Then,by calculating the zero crossing rate of the horizontal and vertical direction of the image, edge preserving filter is used to measure the edge information of image and accurately extract edge feature of image. Finally,by using the extracted pixel features and edge features,defining feature fusion function, and introducing the particle swarm optimization( PSO) to optimize the parameters of the feature fusion mod-el,the ability of insight into image blurring and ringing is improved,image quality evaluation model is con-structed based on fusion feature. Compared with the current no-reference quality evaluation algorithm,the proposed algorithm can effectively evaluate the quality of JPEG,JPEG2000,BLUR distortion images and the evaluation index CC( Correlation Coefficient) and SROCC( Spearman Rank-order Correlation Coefficient) reach 0. 9477 and 0. 9153, respectively. The algorithm has good consistency with the subjective evalua-tion method and can be applied in multiple types of distortion image quality evaluation.