The Research Of Corner Detection Algorithm Based Upon Edges Contour

Corners contain the significant information of an image, and it plays an important role in different applications. The corner detection is often used in computer vision, 3-D reconstruction, target tracking, face recognition and stereo matching, etc. Therefore, the accuracy of the corner detection algorithm is very important.Three kinds of corner detection methods are analysised in this paper: intensity-based methods, contour-based methods and template-based methods. These three methods have their own advantages and disadvantages, exposed the different problems in practical application. Three kinds of improved contour-based methods are proposed in this paper afer analysing their disadvantages. Firstly, proposed two new corners measure based on the gradient information of neighbor pixel at each edge contour. Under different affine transformation, the proposed algorithm compared with several typical algorithms. The result shows that the proposed detector is more competitive with respect to detection accuracy, affine transforms and noise robustness. According to the some problems of Canny detection, a new corner detection based on coarse contour is proposed in this paper, effectively solve the problems of highly dependent on the edge and sensitive to the noise.This topic is completed in the high resolution edge detection of based on multi-scale anisotropic Gaussian kernels national nature science foundation of china. Matlab is used as the software. This methods performance evaluation based upon affine transform for special image database for hundreds of times experiment contrast. The experimental results show that the proposed detector is more competitive with other corner detections and have good robustness.