Visual filling-in and surface property reconstruction
FacultiesFakultät für Ingenieurwissenschaften und Informatik
We study the problem of the reconstruction of achromatic surface properties, namely brightness. We show that the filling-in of brightness through a diffusion process can be linked to the general framework of feature reconstruction through regularization theory that is widely used in computer vision. In particular, it will be shown that brightness filling-in is a means of reconstructing smooth information from local contrast data that minimizes first order derivative information. Our investigation provides a formal link between modeling perceptual data for biological vision and the mathematical frameworks of regularization and linear spatially variant diffusion. The present analysis of filling-in also suggests a new diffusion mechanism that effectively employs confidence information. Simulations on psychophysical stimuli illustrate the mechanism´s potential.
Subject HeadingsComputervision [GND]
Brightness perception [LCSH]
Image processing [LCSH]
Information display systems [LCSH]