Modeling and validation of a psychovisually based image quality evaluator for DCT-based compression

摘要

In this paper, we propose an algorithm for evaluating the quality of JPEG compressed images, called the psychovisually based image quality evaluator (PIQE), which measures the severity of artifacts produced by JPEG compression. The PIQE evaluates the image quality using two psychovisually based fidelity criteria: blockiness and similarity. The blockiness is an index that measures the patterned square artifact created as a by-product of the lossy DCT-based compression technique used by JPEG and MPEG. The similarity measures the perceivable detail remaining after compression. The blockiness and similarity are combined into a single PIQE index used to assess quality. The PIQE model is tuned by using subjective assessment results of five subjects evaluating six sets of images. To demonstrate the robustness of the model, a set of validation experiments is conducted by repeating the subject assessment procedure with four new subjects evaluating five new image sets. The PIQE model is most accurate when the JPEG quantization factor is in the range for which JPEG compression is most effective.