Purchase this article with an account.
Che-Chun Su, Alan Bovik, Lawrence Cormack; Natural scene statistics of color and range. Journal of Vision 2011;11(11):1190. doi: 10.1167/11.11.1190.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Depth (range and changes in range) is an extremely important aspect of the environment that must be recovered from image data. Yet there has been relatively little work done on analyzing the statistical relationships between luminance, chrominance, and range, presumably due to the difficulty in getting co-registered ground-truth range data and RGB pixel data of specific natural scenes. We used a RIEGL laser scanner mounted with a Nikon D700 digital camera and a translating mount to acquire stereoscopic RGB images with co-registered range maps. These images were transformed into the more perceptually relevant CIE L*a*b* color space, and were then encoded by Gabor filter banks with different scales and orientations to roughly capture the kind of information available at the level of primary visual cortex. We examined the conditional distributions relating the luminance or chrominance information with range gradients. Of more relevance to perception, the distributions of range conditioned on the Gabor responses, whether luminance or chrominance, had very similar exponential shapes. We also examined the variations of statistical measures, e.g. mean, standard deviation, and entropy, of range gradients with the changes of the Gabor responses. Most importantly, we found that the depth difference between neighboring pixels increases as the corresponding magnitude of Gabor responses rises. Therefore, the way these range distributions changed as function of Gabor responses indicates that the visual system could, in principle, use these conditional statistics to help recover depth information from the environment. Moreover, these statistical relationships cannot only yield insight into how 3D structure in the environment might be recovered from image data, but may also be applied to various image and video engineering applications, e.g. image de-noising and restoration, and quality assessment of 3D images and video.
This PDF is available to Subscribers Only