Research On Image Corner Detection Algorithm

Corner detection is the basic problem in the field of image processing, which is an important method of low level image processing. Corners are important image features that may be used at subsequent levels of computer vision such as object recognition, stereo matching and motion estimation. Based on B spline scale space, this paper proposes curvature product corner detection algorithm and evaluates the algorithm in terms of corner evaluation criterion.This paper introduces B-spline curvature scale space framework which is constructed in terms of scale space theory and a theory of curve presentation and evolution; moreover, the good properties of curve presentation based on B-spline are showed. Subsequently, in order to obtain better corner detection result, multi-scale product is proposed, and at the same time, the validity is demonstrated through the experiments. So it is introduced into corner detection algorithm of this paper. This paper presents a multiscale curvature product corner detection technique in the framework of B-spline curvature scale space. A scale product function is established derived from the curvature product of the contour at different scales. Local extremes of curvature product at different scales are reported as corners when the curvature product exceeds a threshold.At the same time, in order to show the validity of proposed algorithm. CNN and ACU are introduced. According to the criterions, the comparisons between the proposed approach and other corner detections mentioned in this paper shows that our approach achieves the highest corner number consistency and accuracy. Moreover, the proposed algorithm achieves the better results and much better detection evaluation. Experiments also demonstrate that proposed method has more robustness to boundary noise and trivial details.