Corner Detection Based on Multi-Feature
多特征复合的角点提取方法

The corner is an important local feature of image. To avoid the disadvantages of using the single feature to detect the corner points, a new algorithm based on multi feature is proposed in this paper. In this algorithm, intensity feature and edge feature are contributed to corner detection. First, a fast adaptive SUSAN principle, which utilizes the local gray level feature directly, is proposed for detecting the candidate corners. This improved method can detect features, such as corners, edges and intersections, in different contrast image automatically. For detecting the corners on blurry edges, the candidate corners would include some edge points as a result of reducing the detection threshold. These candidate corners, which include true corners, some edge points and a few false points, are arrayed along the boundary trend by the method of edge element. Through these arrayed points, the angles between approximate straight edge lines are calculated to be as the criterion of determining a corner. Those edge points are removed since they have not significant discontinuous changes in the direction of boundary, i.e. the angles of them are not acute enough, and the false corners due to quantization also are removed by our method. After these steps, the true corners are reserved. The experimental results showed this corner detection method having good capabilities of detection and localization in different contrast image.