Computational visual sensitivity models - a review

The Human Visual Systems (HVSs) is imperfect and contains highly selective visual acquisition sensors. Not all useful information presented to human eyes can be perceived. Generally, a technique to determine whether a signal can be perceived and how well it can be perceived is called human visual sensitivity analysis. More than one hundred years of psychophysical research on HVSs has revealed that human visual sensitivity is not only determined by local characteristics of visual contents (luminance, contrast orientation, spatial and temporal frequency), but also global modulatory factors (visual attention and motion suppression). In this paper, we provide a review on various factors that affect human visual sensitivity and on various computational models for human visual sensitivity. A comparative study on the performance of various visual sensitivity models by simulations is also reported in the paper. Subjective evaluation on noise-embedded video sequences confirms that the introduction of global modulatory factors does improve the performance of the Just Noticeable Difference (JND) profile used in noise shaping.