An automated region growing algorithm for segmentation of texture regions in SAR images

Abstract

Synthetic Aperture Radar (SAR) images have a high geomorphologic information content. Due to the particular operation of this sensor the geomorphologic features of the Earth's surface are enhanced, hence providing valuable detail related to texture of terrain. In this work, speckle reduction in SAR images is considered first by applying selected filters to the original image. The effectiveness of the filtering was qualitatively evaluated for the whole image, and quantitatively estimated on selected smooth and rough texture areas. The quantitative evaluation was done by calculation of image parameters such as contrast, edge sharpening and speckle reduction. This analysis indicated that the geometric filter performs the best for SAR images. The filtered image was input into an automated region growing algorithm that produced segmented texture objects. The texture model and the parameters of the algorithm are based upon the co-occurrence matrix. From this, a new texture distance is introduced. This distance is used to quantify the optimal parameters of the automated algorithm and to determine the effectiveness of the segmentation. Once the image is segmented the morphology of texture regions is obtained, this comprises the following parameters: area, perimeter and fractal dimension. A texture distance between regions is also calculated in order to estimate the texture separation of these regions. A detailed example is provided to illustrate the methodology proposed in this research.