Approximation schemes for image fusion and analysis

Thomas Kaempke
Forschungsinstitut fur anwendungsorientierte
Wissensverarbeitung FAW, Helmholtzstr. 16, 89081 Ulm, Germany Two approximation schemes for computer vision are presented. Fusing registered (spatially adjusted) images is organized along an undirected Markov field resulting in a finer pixel grading of a single image (high resolution image) compared to the original images. Gibbs sampling based on different neighbourhoods approximates the high resolution image. Illustrations show natural and artifical scenes.

A fast deterministic signal approximation scheme is given for image segmentation and feature selection from a given portfolio. Discrete versions of scale spaces are computed by dynamic programming. Variations of the approximation cover acoustic quality control and reading biochip images.