Automatic target recognition in IR(infrared) images is a problem of considerable practical interest in many fields. Specially, segmentation of IR images is very important in the recognition system.
Because of the complexity and changeability of IR images, it is a general way to use the shape of the target in the recognition stage. So we must extract the shape of the target perfectly in the segmentation stage.
In this paper, we propose another region based segmentation scheme in IR images. This scheme is focused on the extraction of the precise shape. After reduction of the entire images, we extract the brightest target region -the seed region. Then it is iteratively expanded to the subregions connected with the seed region. During the region growing, the partial derivatives at the border of the grown region are obtained and the border pels of high edge value are marked. The subregions connected with the marked pels are not selected in region growing.
It is shown that the proposed scheme provides better segmentation than the conventional scheme.
