Edge detection of manmade objects using wavelets in high resolution satellite images

The edge detection problem of man made objects has traditionally been addressed with the use of the Canny and Hough transforms. In recent years considerable interest was developed in new transforms that address the problem of edge detection, especially in case of high resolution satellite images. They are based on the wavelet transform. The challenge is to choose the appropriate wavelet for a particular application which is not known a priori. In this paper, two methods based on wavelets have been introduced and implemented for extracting edges. The first method was based on the biorthogonal mother wavelet while the second method was based on a new wavelet named contourlet. Contourlets have been appropriately designed to locate the edges of an image, thus having a significant advantage compared to classical wavelets for edge detection. The evaluation was made using high resolution images containing specific features with a variety of geometric shapes, in order to understand better the advantages of the new wavelet transform. The results have shown that the wavelet transform using the biorthogonal wavelet produced accurate edge detection results on high resolution satellite images of urban areas Moreover, the contourlet gave very good results, in detecting roads, some of their types, and other linear features.