Subjective Evaluation of Retinal-Dependent Image Degradations

摘要

To understand the contributions and the requirements of image resolution at different eccentricities to perceived image quality, we measured subjective image quality for degraded multiresolution images. A vision model equation was used to estimate contrast threshold as a function of retinal eccentricity. Two high-resolution images were processed using a vision-performance-based algorithm. Eight such degraded images were computed for each original scene by scaling the model-determined performance. During this experiment, the participants were asked to fixate on a target. The images were displayed for 250 ms in each trial to prevent saccadic eye movements. Five observers participated in the experiment and were instructed to rate the image quality on a ratio scale. Each image presentation was repeated 20 times in a random sequence. The average response over the five observers showed that the perceptible image degradation level was slightly higher than predicted by the model equation. This is consistent with our expectations as the visibility of the degradation should be masked by the background image. This experiment demonstrates that the vision model, which is based on visual sensitivity to simple grating patches at different eccentricities and spatial frequencies, provides a useful tool to predict the perceptible image degradation in real complex scenes as a function of retinal eccentricity.