Full Reference Image Quality Assessment by CNN Feature Maps and Visual Saliency
Yu Iwashima, Ji Wang, Yoshiyuki Yashima · 2019
In this paper, we propose a new full reference image quality estimation method by feature maps which are intermediate layer's outputs in convolutional neural network. The novelty of the proposed method is to combine a saliency map reflecting human gaze area with the feature maps of convolutional neural network. In addition, we analyze in detail which layer's feature maps are effective for image quality estimation. Experiments using CID:IQ data set are performed on VGG16 and VGG19 which are deep neural networks for object recognition, and the results show that the image quality estimation accuracy can be significantly improved compared to the conventional method by introducing the saliency map.