Model-based 2D edge detection using bottom-up strategy

Summary form only given. It has been shown that a bottom-up strategy using an improved version of the optimized 2D edge filter fits the requirements of edge detection in real-world 2D images better than the top-down approach. The results are obtained by first operating on the full resolution, gradually restricting it to improve detection capabilities. The 2D filter has been applied to a bottom-up multiresolution edge detection scheme. All contour segments are registered. The behavior is observed while a fine-to-coarse tracing is performed, and the segments are classified as to whether they carry relevant information or not. The optimal edge detector has been tested using two strategies: a top-down procedure and the bottom-up scale-space scheme. While the computational burden is the same in both cases, the bottom-up approach provides considerable improvement in noise suppression and always detects the full length of a contour as obtained at the finest scale.<
>