A Multi-metrics Integrated FR IQA Method Based on Machine Learning

In this paper, a FR (full reference) image quality prediction model learned by AdaBoost BP neural network is proposed by using a bag of features. Firstly some FR image quality assessment methods to improve metrics by considering human visual system are summarized. Further, a group of FR IQA (image quality assessment) metrics are extracted as the candidates of the representation of images. AdaBoost algorithm is finally used to enhance BP network built on a bag of features (9 chosen metrics) for image quality prediction. Then the learned prediction model can map a bag of features into an image quality score as MOS or DMOS. Our experiments are executed on TID2008, TID2013 and LIVE Dataset. Experimental results show that our proposed multi-metrics integrated prediction model works well and get better results than any single metric under complex situations.