In this paper, a no-reference image quality assessment (NR-IQA) algorithm based on a two-stage non-parametric framework is presented. At the first stage, the type of distortion affecting the test image patches is first identified via a nearest-neighbor (NN) based classifier. Utilizing the differential mean opinion score (DMOS) values associated with the labelled patches within the identified distortion class, the quality of each test patch is then predicted using k-NN regression. The predicted scores are then pooled together to obtain the quality score of the test image. The proposed algorithm is simple yet effective. No training phase is required and the algorithm also offers prediction of a local region's quality which is not available in most of the previous NR-IQA methods. Experimental results on the standard LIVE IQA database indicate that the proposed algorithm correlates highly with human perceptual measures and deliver competitive performance to state-of-the-art NR-IQA algorithms.
|Published - Nov 2015
|3rd IAPR Asian Conference on Pattern Recognition (ACPR) - Kuala Lumpur
Duration: 1 Nov 2015 → …
|3rd IAPR Asian Conference on Pattern Recognition (ACPR)
|1/11/15 → …