Blind quality assessment for contrast changed images
Dixiu Zhong, Ping An Shi, Da Pan, Ming Hou, Zefeng Ying, Mingliang Han · 2017 IEEE 3rd Information Technology and Mechatronics Engineering Conference (ITOEC) · 2017
Contrast of image plays an important role in image perception quality and is also susceptive to various factors during image acquisition process. However, only a few image quality evaluation algorithms have been focused on the contrast-changed image quality assessment (IQA), and none of these methods belongs to blind IQA algorithms. Therefore, they cannot be applied to the case when the reference image is not available. Based on the fashionable convolution neural network (CNN), this paper presents a blind contrast-changed image quality evaluation method. First, the distortion image is cropped into patches which are prepared to feed into the network. Then image features used for quality score prediction are extracted by the convolution layers. Finally, a regression layer is applied to map the image features to the space of quality score. Our experimental results suggest that the proposed method is well correlated with subjective evaluations of contrast-changed image quality.