Detection and localization of edge contours

Edge contour extraction plays an important role in computer vision because edge contours are relatively invariant to the changes of illumination conditions, sensor characteristics, etc. In particular, edge contours can be used as matching primitives for correspondence determination, an important step in video geo-registration. In this paper, we present a new approach for edge contour extraction based on a three-step procedure that using a RCBS-based scheme, inherently more accurate results can be produced, even though the edge model used for edges is relatively simple. We also present recursive filters that can efficiently smooth splines by approximating a signal with a complete set of coefficients subject to certain regularisation constraints. We demonstrate our method on both synthetic and real images