A new no-reference quality metric for JPEG2000 images

Abstract

A new technique is developed for no-reference (NR) quality assessment (QA) of JPEG2000 images. Based on the basic activity of general pixels, the proposed measure overcomes the limitations imposed by feature/structure extraction of distorted images. Besides, its structural content-weighted pooling approach involves no parameter and avoids additional procedures and training data for parameter determination. The effectiveness of the proposed measure is demonstrated through its close correlation with subjective quality scores. Besides, the proposed NR quality measure is shown better than the full-reference (FR) quality measure of peak signal-to-noise ratio (PSNR), comparable to the FR quality measure of structural similarity (SSIM) index, and quite competitive among the state-of-the-art NR quality measures. With satisfactory performance at reasonably computational expense and with ease of implementation, the proposed technique is proven to be a quality metric of choice for JPEG2000 images.