Blind Quality Assessment of Camera Images Based on Low-Level and High-Level Statistical Features
Yutao Liu,Ke Gu,Shiqi Wang,Debin Zhao,Wen Gao
DOI: https://doi.org/10.1109/tmm.2018.2849602
IF: 7.3
2019-01-01
IEEE Transactions on Multimedia
Abstract:Camera images in reality are easily affected by various distortions, such as blur, noise, blockiness, and the like, which damage the quality of images. The complexity of distortions in camera images raises significant challenge for precisely predicting their perceptual quality. In this paper, we present an image quality assessment (IQA) approach that aims to solve this challenging problem to some extent. In the proposed method, we first extract the low-level and high-level statistical features, which can capture the quality degradations effectively. On the one hand, the first kind of statistical features are extracted from the locally mean subtracted and contrast normalized coefficients, which denote the low-level features in the early human vision. On the other hand, the recently proposed brain theory and neuroscience, especially the free-energy principle, reveal that the human brain tries to explain its encountered visual scenes through an inner creative model, with which the brain can produce the projection for the image. Then, the quality of perceptions can be reflected by the divergence between the image and its brain projection. Based on this, we extract the second type of features from the brain perception mechanism, which represent the high-level features. The low-level and high-level statistical features can play a complementary role in quality prediction. After feature extraction, we design a neural network to integrate all the features and convert them to the final quality score. Extensive tests performed on two real camera image datasets prove the validity of our method and its advantageous predicting ability over the competitive IQA models.