Research And Implementation On Fast Corner Detection Algorithms Based On MIC

As one of the image features, corner includes abundant image information. Corner detection plays an important role in image processing. Among corner detectors, Harris algorithm has a good performance. But it requires to compute image gradient and three times Gaussian filtering, there is so much operation that it is hardly suit to image real-time applications. SUSAN algorithm does not have to compute derivative, so it is fast. It is very suit to detect the images which have clear edges, but the result is not so good when it comes to the blur images. MIC algorithm, which uses multigrid algorithm, is very fast and performs well in image processing. However, it has a worse efficiency, so the paper bases on the algorithm to do further research for pursuing a good and fast corner detection algorithm.The paper mainly analyzes the traditional corner detectors such as Harris algorithm, SUSAN algorithm. After learning the principle of these algorithms realization, the paper conducts experiment to verify these algorithms. According to the results, the paper summarizes the characteristics of these algorithms. Then the paper focuses on the templates of MIC algorithm and analyzes how these templates affect the results. Combined with the results of these studies, according to the defect that MIC algorithm is hardly to distinguish some points on the edges and corners, this paper presents an adaptive algorithm which can filter some points on the edges. On the clear edges, the method of calculating USAN area is used to filter these points; on the fuzzy edges, a larger template interpolation method is used to calculate corner respond function value in kernel’s neighborhood. Experiments show this algorithm is more efficient in distinguishing some points on the edges and corners, and can achieve the purpose of detecting corners.During the X-corner detection research, the paper use MIC algorithm to accelerate the speed of corner detection. After using MIC algorithm to detect candidate corners, it will be further to implement X-corners localization. The paper adopts two kinds of methods to locating X-corners. One is SV algorithm which has small calculation and is easy to execute, the other is the algorithm based on checkerboard features. The former detects X-corners through symmetry and variance. The latter makes use of the features in the neighborhood of X-corner to detect X-corners. Experiments show that the algorithms can detect X-corners effectively and fast, so they are suitable for X-corner real-time detection.The paper establishes an image corner detection platform which can accurately detect corners in common images and X-corners in checkerboard images. The platform provides a variety of corner detection algorithms, so it can choose a different corner detector to process image according to actual circumstances.