Adaptive machine learning algorithm for multispectral image analysis

Abstract

This paper describes a machine learning algorithm for analyzing the multispectral images of natural scenes. A mathematical training algorithm has been developed to guide the operation of a statistical pattern recognition technique for detecting and extracting the image clusters in a multidimensional feature space. For this purpose, the peak modality of 1-D image histograms is selected as the mathematical training criterion. The algorithm is applied to the clusters of the color images of natural scenes in 3-D feature space. During the training process, image clusters are detected in some well-defined decision elements using constant lightness and chromaticity loci of the uniform color space. This gives non-parametric estimates of the clusters' distributions without imposing any constraints in their forms. The linear discriminant method is then used to project simultaneously the detected clusters onto a line for region isolation. This permits utilization of all the spectral properties for object recognition and inherently recognizes their respective cross correlation.© (1992) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.