Real Time System for Multi-Sensor Image Analysis through Pyramidal Segmentation

摘要

A state of the art, fully functional, multi-scale and multi-channel segmentationtool has been developed. It is based on the recently developed computational theory of the 2-normal segmentations. A fast multi-scale pyramidal algorithm has been designed and applied to the theoretical variational segmentation model of Mumford-Shah. This algorithm has a multi-channel capability, as well as a much more general class of solutions. Namely, a piecewise polynomial segmentation is natural to the pyramidal multi-channel framework. The piecewise affine segmentation has been implemented and tested. Application specific channels include: gray scale information, two-dimensional wavelet channels for texture discrimination, and multi-scale singular feature channels. The accuracy of the pyramidal segmentation algorithm has been experimentally compared to the accuracy of two other modern segmentation algorithms. The performance of the pyramidal algorithm has shown an average four-fold reduction in error measurements. Computational experiments with reconnaissance photography, multi-sensor satellite imagery, medical CT and MRI multi-band data have shown a great practical potential of this novel technique. Preliminary experimentation in clutter removal via multi-channel segmentation points to a totally new class of feature preserving decluttering algorithms.