Edge Detection Based on Fast Adaptive Mean Shift Algorithm

Edge detection is arguably the most important operation in low level computer vision. Mean shift is an effective iterative algorithm widely used in edge detection. But the cost of computation prohibits Mean shift algorithm for high dimensions feature space. In this paper, a fast adaptive mean shift algorithm is proposed for edge detection. It makes use of one approximate nearest neighbors search method, i.e. LSH (locality-sensitive hashing) firstly, which dramatically reduces the computation of iterations in high dimensions. Moreover, the LSH procedure can help to determine the bandwidth of the kernel window adaptively. The experimental results show the effectiveness of the proposed approach.