A novel perceptual image quality measure for block based image compression

The challenge of finding a reliable, real-time, automatic perceptual evaluation of image quality has been tackled continuously by researchers worldwide. Existing methods often have high complexity, or are dependent on setup specifics such as image size, or else have low correlation with subjective quality. We propose a novel, easy to compute, image quality score which reliably measures artifacts introduced in block based coding schemes. The proposed score, named BBCQ (Block Based Coding Quality) lies in the range 0-1 with 1 indicating identical images, and is composed of three components. These components are based on a pixel-wise error using PSNR, evaluation of added artifactual edges along coding block boundaries and a measure of the texture distortion. These three measures are calculated on image tiles, whose size depends on image resolution, and are combined using a weighted geometric average. The obtained local scores, one per image tile, may then be used for local quality evaluation, or pooled into a single overall image quality score. The proposed quality score enables reliable, real-time, automatic perceptual evaluation of the quality of block-based coded images. BBCQ has been successfully integrated into an automatic, perceptually lossless, JPEG recompression system.