Object searching in scale-space

One of the important problems in multi-scale analysis is the detection of feature evolution in scale-space. Using the Laplacian of 2-D Gaussian as the kernel, the detected zero-crossing of a corner in scale-space is an ellipse whose radius of curvature is determined by scale analysis and the slope of the corner's boundary. Based on this result, we propose an object searching method for meso-scale eddy detection in remotely sensed ocean images. The method is based on the strategy of detecting possible eddies from large regions to small regions. Instead of applying the Hough transform directly to the edge map, we first search the position of the eddies by using an ellipse fitting. Then we apply the local Hough transform to each of the candidate sub-regions to detect the regions which have no eddy inside. Finally, we use image segmentation and object recognition techniques to detect the precise positions of the eddies.