The multiple interaction model for nonshallow scatterers buried beneath 2-d random rough surfaces

It is now clear that natural texture images can be discriminated by multifractal exponents. Most natural images, such as geographical images , are all textural in nature. In remote sensing images, different regions possess different texture and have different multifractal exponents. These properties are thus ideal for use in image segmentation. Like fractal dimensions, multifractal exponents are rotation invariant and intensity invariant. As these are the inherent properties of most natural regions and objects, multifractal exponents can be used to accurately describe these images. In this simulation, four multifractal exponents (q = 01; 0; 1; 2) are chosen to segment a synthetic aperture radar (SAR) image. In order to highlight the edges, the K-means clustering approach [7], which is a fast and appropriate cluster technique for large data sets, is also applied. Using four multifractal exponents and applying the K-means method, the cluster results can be obtained. Finally, the results are mapped from the feature space to the image space. The image we used in this paper is as shown in Fig. 2, which is a X-band SAR image spanning an area of about 100 kilometers by 50 kilometers. This is an image of the area surrounding the city of New Orleans, LA in the southeastern U.S. The image size is 520 2 450 pixels and the dark area in the center is Lake Pontchartrain. The thin line running across the lake is a causeway connecting New Orleans to the city of Mandeville. Segmentation of the image is performed on workstation running at 250 MHz. Four multifractal exponents are estimated for each pixel based on a 15 2 15 image window. By using the K-means method, each pixel is classified into two groups: land or lake. The result obtained using the proposed method is shown in Fig. 3 in which the boundaries of Lake Pontchartrain are shown clearly and correctly, even the causeway. As multifractal exponents reflect the spatial statistics of re-gions/objects, they can provide a good description of the geographical image and lead to good segmentation results. V. CONCLUSION In this paper, we present a novel multifractal estimation method to describe the local scaling properties of a region/object. Computer experiments show that the proposed method performs well at different window sizes and different gray-scale levels. The proposed method has also been shown to apply successfully to the segmentation of geographical images. REFERENCES [1] B. R. Kerman, " Information states …