Regional segmentaion of multispectral image using local texture/spectral joint histgram

We present a new method for segmentation of multispectral image. The method can be divided into three steps. Firstly, the image is partitioned into small regions using mean shift algorithm; then texture-spectral joint histograms are extracted from those regions; thirdly, an adaptive, multiscale merging process is defined to get final result. Experiments on several synthetic images show that the method is superior to two competitors.