结合美学特征的无参考色调映射图像质量评价

如何对色调映射图像(Tone-mapped Image, TMI)质量进行有效评价是图像处理领域亟待解决的问题. 现有方法仅是低动态范围图像质量评价方法的改进, 并未考虑TM图像本身独有的特点. 基于此, 本文提出了一种结合美学特征的TMI质量评价方法. 从颜色保真度、色温分布、对比度、黑暗度、自然度5个方面出发提取图像的美学特征, 并在梯度域、张量域、高斯差分域和拉普拉斯域提取变换域特征, 利用支持向量回归训练得到质量评价模型. 在ESPL-LIVE HDR库上的实验表明, 现有代表性方法的Spearman秩相关系数最高为0.752, 而所提出模型可达0.766, 与人眼感知结果具有更强的一致性.

No-reference image quality evaluation of tone-mapped images based on aesthetic characteristics

How to effectively evaluate the Tone-Mapped Image (TMI) has become much-needed subject worth probing in the field of image processing. The existing method offers only improvement of the low dynamic range image quality evaluation method without considering the unique characteristics of the TMI itself. To tackle this weakness, in this paper a TMI quality evaluation method is proposed which takes the aesthetic features into consideration. The aesthetic features of the image are extracted from five indices: color fidelity, color temperature distribution, contrast, darkness, and naturalness. The transform domain features are extracted from the domains of the gradient, tensor, difference of Gaussian, and Laplacian. Based on the feature extraction, the quality evaluation model is trained through support vector regression. Experimental results by applying to the ESPL-LIVE HDR Image Database show that the Spearman rank correlation coefficient of the proposed model can reach 0.766 comparing with the other existing methods being no greater than 0.752, which indicates that the proposed model is in better agreement with the results given by human eye perception.