Self-adaptative variable-metric feature point extraction method

A feature point extraction method for self-adaptative variable-metric constructing image pyramid is proposed to accelerate the feature matching. In this method, number of FAST feature points is adopted as information content quantization in scale space representation and pyramid hierarchy is carried out according to the information difference of blurred images in the neighboring layers. By adjusting scale parameters, Uniform change of detail feature in neighboring images is realized, number threshold of matching points is used to control the height of pyramid and matching efficiency is improved by applying matching instruction strategy named “matching and constructing at the same time”. Last, The contrast experiment is implemented between proposed method and three detection methods-SIFT, FAST, and ASIFT. The experiment results indicate that correct matching rate of the method can reach 43.59% under various scales. It increase by 25.51% compared with SIFT. Feature points can still show the targets correctly after they underwent all kinds of changes in lights and angles. The method referred to in the paper selects parameters adaptively according to the feature of target image. It can obtain ideal matching effects without manual adjustment and adapt to feature extraction and matching in various changeable conditions in high efficiency.