Abstract
No Reference Image Quality Assessment (NR-IQA) refers to automatic quality assessment of an image using an algorithm such that the only information that the algorithm receives is the distorted image. Moreover, the general purpose NR-IQA algorithms predicts the quality of image with no information about the distortion that affecting the tested image. BIQI framework [1] classifies the distortions in the distorted image first, and then followed by distortion-specific quality assessment method to predict the final quality score. In this paper, we improve the classification portion by investigating the performance of different classification techniques and different features. Each features in our features collection is evaluated using different features evaluation techniques. We constructed a set of optimal features that classify image distortions with a high accuracy rate. Furthermore, we find that the best performing classifier is multiclass classifier (Exhaustive Correction Code) with logistic regression as base classifier. Our evaluation shows that our trained model outperforms state-of-the-art classification methods.