关键词: 图像质量评价, 部分参考型, 模糊分类

Abstract:

Image quality assessment is an important branch in the fields of image processing. It would be employed for calibrating image processing system or algorithms, and be applied for algorithm optimizing and parameter setting. Reduced-reference image quality assessment (RR-IQA) has become to be one of the focuses in image processing fields. Inspired by the fuzzy human evaluation, an efficient RR-IQA framework is proposed in this paper. In the framework, the images are allocated into several fuzzy sets with their degrees of memberships. The natural scene statistics (NSS) in wavelet domain is used for extracting features. After that, a multi-class fuzzy classifier is training for assigning image features into fuzzy sets with their corresponding degrees of memberships. Contrast to the typical RR-IQA methods, the proposed one relates well with the human evaluations and has low computational complexity. The experimental results demonstrate that the proposed framework outperforms the state-of-the-art reduced-reference methods.

Key words: image quality assessment, reduced-reference, fuzzy classification