QUALITY ASSESSMENT MODEL FOR SMARTPHONE CAMERA PHOTO BASED ON INCEPTION NETWORK WITH RESIDUAL MODULE AND BATCH NORMALIZATION

The popularity of smartphones has made it increasingly common to take photos with smartphones. For those who design and develop cameras, as well as those who use cameras, it is advantageous to have a way to assess the image quality of a smartphone camera. On account of the distortion of pictures taken by smartphones is different from that of traditional pictures, traditional methods of image quality assessment (IQA) cannot be directly applied to pictures taken by smartphones. In this paper, we submit four models for quality assessment of photos taken by smartphones. We use a pre-trained saliency prediction model SalGAN to preprocess data, and extract different features of the image for different indicators such as exposure, noise, texture, color. Then we input them to the modified Inception network with residual module and batch normalization for training. Our models outperform traditional no-reference IQA methods on the training set. The average SROCC reaches 0.45, 0.36, 0.33, 0.36 for exposure, color, noise, texture respectively.