The present invention relates to a method for extracting a corner point that can represent a feature in an outline image, to obtain an outline image by processing a preprocessing process from a raw image obtained from a camera, multi-layer scale decomposition by wavelet filtering the obtained image And curvature measurement, search the wavelet local maximum value in the decomposed curvature signal, construct a scale space map with the searched local maximum value, and construct a local maximum chain with the evolutionary characteristics of the wavelet maximum value. After the chain search, the corner candidate point is generated, and the candidates found are subjected to Lipps's normality test to remove the edge noise and unnecessary surplus corner candidate points to extract the final corner points. Compared to the scale method, multilayer scale decomposition and curvature measurement are performed in one step. It is reduced and the improvement in performance for the quantization noise and contour noise does not cause an unnecessary excess corner points.
展开▼
