基于特征域奇异值分解的图像质量评价

Image quality assessment based on singular value decomposition in multiple feature domains

为了克服传统图像质量评价算法泛化能力不足的问题,提出一种基于特征域奇异值分解的图像质量预测模型.首先从多个特征域(图像及其梯度和相位一致性)中分别比较图像局部的奇异向量和奇异值差异完成视觉特征提取,随后利用支持向量机完成图像感知质量预测.实验表明:所提出的基于支持向量机而构建图像质量预测模型不仅在单个图像数据库上的表现要优于传统的图像质量评价算法,而且有着良好的跨数据库性能变现,表现出较高的泛化性;通过用集成学习器取代单个支持向量机,图像感知质量预测模型的泛化能力还可以进一步提高.

To solve the insufficient generalization ability of the traditional image quality assessment （IQA） algorithms, an image quality predication （IQP） model based on the singular value decomposition in multiple feature domains was proposed. The visual features were extracted by comparing the difference of sin gular values and singular vectors between the corresponding local neighborhoods of reference and test images in the multiple feature domains （images and their gradient and phase congruency maps） , and then fed into a support vector machine （SVM） to predict the perceptual quality of images. Subsequent experiments show that, proposed IQP model built on the SVM not only has a better performance than the traditional IQA algo rithms on individual image databases, but also exhibits good generalization ability by having a good across-im- age-database performance. By replacing the SVM with an ensemble learner, the generalization ability of the proposed IQP model can be improved further.