Research On The Corner Detection Algorithms Of Digital Images

Corner detection is a basic problem in image processing, whose accuracy and efficiency directly affect the performance of the subsequent image processing. As an important feature of image, comer is widely used in image matching, pattern recognition,3d reconstruction and other fields. As corner has much superiority compared with other local invariant features of image and it describe the main information of the gray value change in the images with less information. So the research on comer detection algorithm has great significance.In this thesis, the basic methods of comer detection are firstly studied. Then aiming at the problems existing in the existing comer detection methods, such as the complex corners cannot be located accurately, the complex corners'response values cannot be calculated and the threshold of the comer response value need be set manually, the in-depth research is carried out. The specific research work includes the following aspects:(1) With analyzing the relative researches, we discuss the corner's evaluation standards and analyze the comer detection algorithms based on edge information and gray information. And we describe the general steps detecting the corner. With the upper analyses we lay the foundations of the research in this thesis.(2) Aiming at the problems that SUSAN comer detection algorithm can't locate the complex corners accurately and not calculate the complex corners'response values, a new detection algorithm based on the annulus template is proposed. Although it has the advantages of simple structure and high efficiency, the SUSAN algorithm's detection capability for complex comer is insufficient. Therefore, some scholars proposed that it is an effective way to improve the detection ability of SUSAN algorithm for complex comer by additional annulus template. But there are still some shortcomings in this kind of algorithm. In this thesis, the corner response values of complex corners are calculated by transforming the annulus template with the circular template in original SUSAN algorithm, so as to overcome the problem that the improved algorithm based on the circle template cannot perform non-maximum suppression; the anti-noise performance of the algorithm is improved by changing the search range of complex comer of the original algorithm and the calculation of the frequency of the transition of gray value on the circle template.(3) Given the response threshold value need to be set manually in the Harris algorithm, the improved algorithm based on adaptive threshold is proposed. The corner response values in Harris algorithm usually have wide ranges and have different ranges for different images, so it is important to realize the adaptive threshold of the corner response value. This thesis proposed an adaptive algorithm regarding the sum of the mathematical expectation and the standard deviation of the corner response values as the threshold, which can supplies the threshold of every pixel automatically.