Study On Fast Segmentation Algorithm Of Multi-dimension SAR Image

As an important link of image procession, image segmentation is a technique which partitions the multi-region image into setting requested regions by segmentation technique. In the paper, an efficient projection-based fast segmentation algorithm for multi-diemnsion SAR image is improved. First, an advanced K-L transformation (weight K-L transformation), providing support for the algorithm in theories, is proposed. Then, by eigen-decomposing, we propose a new method of orthogonal vector and select a proper referenced subspace to be reference the vector of projection of signature vector in the projection algorithm, such as the orthogonal subspace of the mean image signature vector. By projecting the multi-dimension image data into referenced subspace, we transform the multi-dimension image data into one dimension data as one dimension projection length, which degrades the dimensions of the multi-spectral image data, after weight K-L transformation, image is partitioned by a one dimension segmentation technique, such as Moment-preserving algorithm. The projection procedure is performed recursively and the image is partition into setting requested regions according to each terrain characteristics. Final,the N-nearest neighbor algorithm is adopted to merge the image regions according to their spatial-correlation. Simulation results performed on simulated data and measured data in actual demonstrate the efficiency of the proposed algorithm.