A hybrid learning-based framework for blind image quality assessment

Abstract

Blind image quality assessment aims to evaluate the quality of a given image without the availability of the original ground truth image and without the prior knowledge of the types of distortions present. Instead of using hand-crafted features to address specific type of image distortions, a hybrid learning-based blind image quality assessment approach is proposed in this paper to address more challenging mixed types of image distortions. The proposed approach integrates the convolution neural network (CNN) as a feature extractor, plus the support vector regression method to learn a mapping function from the CNN-trained features to the quality score of the input image. Extensive experiments are conducted using both standard image dataset and real-world surveillance video dataset to demonstrate the superior performance of the proposed approach.