Completely Blind Image Quality Assessment Using Latent Quality Factor From Image Local Structure Representation

Although opinion-unaware (OA) blind image quality assessment (BIQA) is the most difficult task, it is still very attractive because of its great potential for good generalization capability and practical usage. In this paper, a novel OA-BIQA algorithm is proposed. This algorithm is based on the construction of a 'quality aware' collection of statistical features based on a simple and successful local structure descriptor generalized local binary pattern (GLBP). GLBP statistical features of images are regarded as the latent characteristics that can distinguish the visual distorted image from those of 'natural' or 'pristine' images. First, the statistical GLBP feature vectors learned from a corpus of the pristine images are divided into different subgroups. Second, a set of multivariate Gaussian (MVG) models are learned from each subgroup features fitting. Finally, the quality of a test image is evaluated by integrating the distances between its pair-wised parameters which describes each MVG models and that of each MVG models learned from a corpus of pristine images. Experimental results show that the proposed model delivers performance correlates well with human difference mean opinion scores on the LIVE IQA database.