Local Spatial Frequency Analysis for Computer Vision
CARNEGIE-MELLON UNIV PITTSBURGH PA ROBOTICS INST
Pagination or Media Count:
A sense of vision is a prerequisite for a robot to function in an unstructured environment. However, real-world scenes contain many interacting phenomena that lead to complex images which are difficult to interpret automatically. Typical computer vision research proceeds by analyzing various effects in isolation e.g. shading, texture, stereo, defocus, usually on images devoid of realistic complicating factors. This leads to specialized algorithms which fail on real-world images. Part of this failure is due to the dichotomy of useful representations for these phenomena. Some effects are best described in the spatial domain, while others are more naturally expressed in frequency. In order to resolve this dichotomy, we present the combined spacefrequency representation which, for each point in an image, shows the spatial frequencies at that point. Within this common representation, we develop a set of simple, natural theories describing phenomena such as texture, shape, aliasing and lens parameters. We show these theories lead to algorithms for shape from texture and for dealiasing image data. The spacefrequency representation should be a key aid in untangling the complex interaction of phenomena in images, allowing automatic understanding of real-world scenes.