PROJECT TITLE :

A Feature-Enriched Completely Blind Image Quality Evaluator - 2015

ABSTRACT:

Existing blind image quality assessment (BIQA) strategies are mostly opinion-aware. They learn regression models from training images with associated human subjective scores to predict the perceptual quality of test pictures. Such opinion-aware strategies, however, need a giant quantity of coaching samples with associated human subjective scores and of a variety of distortion sorts. The BIQA models learned by opinion-aware strategies often have weak generalization capability, hereby limiting their usability in follow. By comparison, opinion-unaware strategies do not need human subjective scores for coaching, and so have larger potential for good generalization capability. Unfortunately, thus far no opinion-unaware BIQA method has shown consistently better quality prediction accuracy than the opinion-aware methods. Here, we tend to aim to develop an opinion-unaware BIQA methodology that can compete with, and maybe outperform, the prevailing opinion-aware ways. By integrating the options of natural image statistics derived from multiple cues, we have a tendency to learn a multivariate Gaussian model of image patches from a assortment of pristine natural images. Using the learned multivariate Gaussian model, a Bhattacharyya-like distance is used to live the quality of every image patch, and then an overall quality score is obtained by average pooling. The proposed BIQA technique will not want any distorted sample pictures nor subjective quality scores for training, nevertheless intensive experiments demonstrate its superior quality-prediction performance to the state-of-the-art opinion-aware BIQA ways. The MATLAB source code of our algorithm is publicly out there at www.comp.polyu.edu.hk/~cslzhang/IQA/ILNIQE/ILNIQE.htm.