Blockwise distortion measure for statistical and structural errors in digital images

摘要

A blockwise distortion measure is proposed for evaluating the visual quality of compressed images. The proposed measure calculates quantitatively how well important visual properties have been preserved in the distorted image. The method consists of three quality factors detecting contrast errors, structural errors, and quantization errors. The proposed method outperforms PQS for a set of test images, and is much simpler to implement. The method should also be applicable to color images; properties like color richness and saturation are captured by the quantization and contrast measures, respectively.