Blind image quality assessment with small training sets by local and global features

• Hua Yang,
• Hao-Rong Li,
• Peng-Jie Wang,
• Cheng-Yang Du,
• Xiao-Dong Bi,
• Hua-Wen Chang,
• Ming-Hui Wang,

Vol. 19, No. 4, April 30, 2025
10.3837/tiis.2025.04.010, Download Paper (Free):

• Image quality assessment (IQA)
• Asymmetric generalized Gaussian distribution (AGGD)
• Curvelet transform (CT)
• Dual-tree complex wavelet transform (DTCWT)

Abstract

Most of the existing blind image quality assessment (BIQA) algorithms require a large number of samples in the training process, however, in many cases, there is not enough training data to support the training of the models. To solve the above problems, we can extract features that are more consistent with the human visual system (HVS) and make the algorithm less dependent on training data. To evaluate the quality of an image, the HVS initially swiftly and automatically builds a global view before progressively focusing on a few small regions. Therefore, in order to more closely simulate perception of the HVS, both local and global (LG) information should be considered when assessing image quality. However, most IQA methods consider only one of the local or global features. The proposal in this paper is to include LG features for BIQA. Two schemes are designed for the feature extraction, which lead to two specific BIQA methods, namely wavelet-based LG-IQA (WLG-IQA) and curvelet-based LG-IQA (CLG-IQA). The proposed methods can simulate the perception characteristics of the HVS to the image quality more effectively, so as to improve the accuracy of model prediction. It is worth mentioning that WLG-IQA and CLG-IQA can also obtain excellent and stable results when few samples are used for training. Experimental results show that the suggested two methods are highly correlated to the subjective scores, and competitive with some existing BIQA methods.

Statistics

Cite this article