Convolutional Neural Networks Based on Residual Block for No-Reference Image Quality Assessment of Smartphone Camera Images

The quality of image captured by smartphone camera is one of the most important factors influencing consumers’ choice of mobile phones. Since the objective evaluation methods specifically designed for the quality assessment of smartphone camera image are relatively rare, it is meaningful to design an effective model for this challenge. In this paper, we propose a carefully-designed Convolutional Neural Network (CNN) with residual block to predict image quality without a reference image. Within the network structure, the feature extraction and regression are integrated into one optimization process. The input of network is selected using the saliency map generated by SalGAN. Experimental results show that the model proposed can obtain a better performance for quality assessment of smartphone images on all four aspects viz. color, exposure, noise and texture than the traditional noreference image quality assessment (NR IQA) methods.