In this paper, we present a new segmentation algorithm based on subband decomposition for SAR images. The decomposition of the SAR images into subband images is performed by 2D four channel non separable filter bank having lattice structure. The four channel filter bank with lattice structure has perfect reconstruction property, while it guarantees good stop band attenuation for each of the analysis filters. The analysis filters as well as synthesis filters are combined into a single structure due to the lattice property. The decomposition and reconstruction levels are accomplished simultaneously using this single structure. Higher order perfect reconstruction 2D lattice filter bank can be obtained from a lower order perfect reconstruction 2D lattice filter bank, simply by adding more lattice sections until the desired frequency specifications are obtained. Once the decomposition of the image is performed, then several different features are extracted from these subband images. A minimum distance classifier is used in the classification step. Results obtained with wavelet transform based segmentation method which use separable filters are also included for comparison purposes.