Abstract
Purpose
Monocular depth cues can provide a vivid sense of depth through image structure (such as perspective and occlusion) or through differential changes with movement (such as motion parallax). Nevertheless, it is unclear how the perception of depth from these cues degrades with resolution, and in particular how it degrades with block pixellation at low resolution.
Depth perception with block pixellation in the low resolution limit is of considerable interest, as it has application to retinal prostheses that are implanted within only one eye. Furthermore, in the case of block pixellation, the high spatial frequency noise of pixellation edges masks low-frequency monocular depth cues, generating the perception of a flat image. Pixellation also generates false depth cues, such as same-size objects (pixels) bordered by parallel lines (pixel edges), thereby indicating a single image depth. Post-pixellation blur removes the false depth cues and unmasks the real depth cues to generate a vivid sense of depth.
Results
Nine naïve subjects rated the depth they perceived in natural images at varying levels of pixellation and blur. Depth ratings of block pixellated images were compared with images of the same resolution that had been block pixellated and Gaussian blurred. Further study showed that even after recognition of an image feature, a significant difference between the depth ratings of pixellated images and those of pixellated and blurred images persisted. Dynamic depth cues such as motion parallax were also studied with motion video, demonstrating similar striking differences.
Conclusions
Depth perception caused by monocular depth perception cues was found to be impaired when the images were pixellated. The impairment was found to be alleviated with the addition of optimal post-pixellation Gaussian blur. Optimization of monocular depth perception in the low-resolution limit may critically improve the functionality of low-resolution visual prosthetic devices (such as intraocular retinal prostheses).
Meeting abstract presented at VSS 2012