Abstract
A major source of redundancy in natural images is the similarity between the left-eye and the right-eye images for a binocular viewer. This redundancy was investigated by analysing the statistics of pairs of pixels drawn from either monocular or binocular images. Joint probability functions were formed for pairs of pixels as a function of their spatial separation. Monocular pixel pairs showed a high degree of redundancy. Mutual information was high for adjacent pixels, and reduced as the separation between the pair increased. When each pixel was drawn from a different member of a binocular image pair, mutual information was lower for nearby pixels than for monocular images, and fell off more gradually with increasing separation. This may be attributed to the effects of binocular disparity. Further, for binocular pairs mutual information depended on image location, being higher and falling off more sharply with separation, for points near fixation. This may be related to the fact that disparity is likely to be smaller for central than for peripheral image locations. It is concluded that two-pixel statistics are useful for understanding the redundancy in binocular images, and the distributions of disparities they contain.
| Original language | English |
|---|---|
| Pages (from-to) | 96-96 |
| Number of pages | 1 |
| Journal | Perception |
| Volume | 33 |
| Issue number | ECVP Abstract Supplement |
| Publication status | Published - 2004 |