Multicovariance Matched Filter for Target Detection in Images.
Final rept. Jul-Dec 94,
BROWN UNIV PROVIDENCE RI LAB FOR ENGINEERING MAN/MACHINE SYSTEMS
Pagination or Media Count:
Our original research on the multicovariance matched filter deals with optimum low resolution target detection in a single-frame, multicolor image, such as a multispectral infrared or polarimetric synthetic aperture radar picture. The multicovariance method completely uses all the joint variability of the problem, in both space and frequency, in a way that generalizes both the traditional spatial matched filter and also techniques involving scalar ratios between frequency bands. The main new focus of our work, directed toward achieving the best target detection performance that is possible, is to develop a preprocessing step involving optimal adaptive estimation of the local clutter background. This involves segmenting the image into regions, which correspond to different backgroundclutter statistical models. Statistics of real data are being studied and used in new state-of-the-art hierarchical segmentation algorithms based on Markov Random Field, polynomial and autoregressive models for vector-valued random processes. The major algorithmic challenges here are in estimating the best possible backgroundclutter models and in accurately estimating the boundaries between different model regions. We are in the process of developing extremely efficient and robust algorithms to estimate these clutter models. These are similar to familiar algorithms from mainstream signal processing, but solve the interpolation problem for Markov Random Fields, which is different than the usual linear prediction problem. MM
- Infrared Detection and Detectors
- Active and Passive Radar Detection and Equipment