A Robust Method For Completely Blind Image Quality Evaluator With Enriched Features

Abstract An important aim of study on the blind image quality assessment (IQA) problem is to devise perceptual model that can predict the quality of distorted images with prior knowledge of the images or their distortions. The existing no reference (NR) IQA algorithms require some knowledge about distortions in the form of training examples and corresponding human opinion scores. This type of quality assessment methods are called as Opinion Aware methods. However we have recently derived a blind IQA model which is Opinion Unaware that does not require human subjective scores for training. Thus, it is ‘completely blind’. The new IQA model, which we call the Natural Image Quality Evaluator (NIQE) is based on the construction of a collection of ‘quality aware’ statistical features based on domain natural scene statistic (NSS) model. We learn a multivariate Gaussian model of image patches by integrating the features of natural image statistics. Using this learned multivariate Gaussian model, a Bhattacharyya-like distance is used to measure the quality of each image patch, and then an overall quality score of the image is obtained by average pooling the qualities of each image patch. Thus opinion-unaware BIQA method has shown consistently better quality prediction accuracy than the opinion-aware method.