A scalar scheme for multispectral images segmentation through multi-thresholding

摘要

In this paper, we present a scalar approach of the multispectral image segmentation. Informations contained in these images are often complementary but present sometimes important redundancies. The first step of processing consists in making a selection of relevant bands. Each of the spectral bands is first characterized by a global histogram built from significant peaks of local histograms. An aggregation by using the global histograms allows then to form a set of bands classes maximizing an entropy criterion. The representative band of each class is that minimizing a dissimilarity measure with the center of the class. Each of the selected bands is then segmented by using a technique of histogram multi-thresholding. This one is achieved by an iterative gray levels aggregation operating on the global histogram. At last, a fusion that combines the multi-thresholding results of the selected bands allows to obtain the final segmentation. This scheme will be illustrated in the frame of an application in high resolution spectral imagery acquired by the CASI (Compact Airborne Spectrographic Imager).