Unsupervised hierarchical SAR image segmentation using lossy data compression

This paper presents a method called hierarchical unsupervised segmentation using lossy data compression for SAR images, where superpixels are used instead of pixels. In the present paper, merging the superpixels is dealt with by combining features such as edges, textures, and brightness. This procedure is done in two stages. The first stage is merging all superpixels until there is no distinct boundary between them. In second stage, merging superpixels is performed if lengths of data codes are minimized under definite distortion. The algorithm has been implemented on SAR images and it was observed that this algorithm has an appropriate accuracy and an acceptable speed.