Abstract: For the problem of improving the measuring efficiency in blade machining process， a new method was proposed for extracting feature points， which was based on the curvature mutation of blade section curve. According to the monotonicity of discrete points integral curve， section curve was classified into the curvature smooth region and curvature mutation region. The basis feature points were selected by the curvature of discrete points and the character of region. B-spline was structured to approximate the original curve. The extraction of feature points was completed when the Hausdorff distance between the constructed B-spline and original curve met preset error value. Calculations and experimental results showed that the density of feature points are related to curvature variation of section curve. When error is set as 0.05mm， the average compression rate can reach 98% and average computation time is 103 seconds. Comparing to the traditional method， it can convergence to preset error value quickly. The proposed method is potentially useful for improving measuring efficiency of blade.

Key words: blade section curve, discrete curvature, basis feature points, curve approximation, Hausdorff distance