A New Vision-Based Evaluation Method for Image Noise: A Computational Subjective Evaluation Method Using A Cooperative Human Vision Model

摘要

A new objective image noise evaluation method based on human vision's perceptual responses has been developed. The purpose of this research is to develop an objective image noise evaluation method that produces results equal to human subjective opinions. Concretely, the following two-step approach, i.e., computational subjective evaluation method, has been proposed: the first is to simulate perceptual responses to image noise by using a cooperative human vision model; the second is to design evaluation criteria based on the model's responses. The proposed method was applied to noise evaluation for halftone patch images with a uniform density outputted from digital color copying machines. The following results were obtained: (1) the vision model simulated correctly perceptual responses to the patch images' noise; (2) two kinds of evaluation criteria designed here correlated closely with the subjective evaluation score (correlation coefficient > 0.96); (3) the quality level of single- colored images was a factor to affect the performance of the copying machines used. These results suggest that the proposed method is effective in producing objective results in good agreement on subjective judgments of humans and that human vision's adaptively-changing perceptual responses are appropriately reflected in the criteria.