Convolutional Neural Network for Blind Image Quality Assessment

Abstract

Blind image quality assessment (BIQA) methods can measure the quality of distorted images even without referencing the original images. This property is indispensable in the image processing field because reference images are normally not available in practice. Unlike the existing trained models, in our work, the training process is constructed as an end-to-end learning mechanism that minimizes the loss between the predicted score and the ground-truth score of the human vision system (HVS). Moreover, a convolutional neural network (CNN) takes distorted images as input and outputs the related score for each image. In this paper, we evaluate the proposed method on six publicly available benchmarks and the cross-database validation performance on the LIVE, CSIQ and TID2013 databases. The experimental results show that our proposed method outperforms other state-of-the-art methods.