dc.description.abstract |
In the 1950s, scientists predicted it would take ten years to develop
robots capable of performing most human tasks. Today, 50 years later, scientists are still predicting it will take ten years. The primary obstacle to building such robots is the problem of interpreting visual data, or image processing. Computers process
images as grids of grey-level values. The structures in those values are the structures in the image. In other words, image processing requires mathematical analysis of images, where the images are represented as grids of numbers.
At the most basic level, images consist of shapes and textures. These textures range from completely regular (e.g., wallpaper) to completely random (e.g., white noise).
The mathematical qualities of regular textures are significantly different from the qualities of random textures because of the high degree of structure present in regular textures and absent in random ones. This project examines structured textures, those close enough to regular textures to be viewed as deformed versions of regular textures (e.g., Figure 1.1). In particular, we are interested in learning how to relate properties of the regular texture with properties of the structured texture (i.e., after deformation).
Ultimately, we want to answer the question: Given an unknown structured texture, can we recover a regular texture and the associated deformation? |
en |