中文摘要：

传统Harris算法依据经验确定兴趣值计算参数,并且对于一些类型的角点识别能力较差。针对这些问题,该文提出了一种基于局部标准差和对数计算的多尺度角点检测方法。该方法通过对数化梯度取降低边缘响应对候选点兴趣值的影响,有效地检测不同类型的角点。并重新定义兴趣值函数,由标准差的统计特性计算兴趣值,避免主观选择参数,使算法具有更高的客观性。实验结果表明,该方法具有检测复杂类型角点、精确定位,并具有旋转、灰度、噪声、尺度不变性。

英文摘要：

Traditional algorithms of Harris needs to decide parameter for computing interest values of pixels by experience and the recognition ability for some types of comers is poor. To solve these problems, this paper proposes a comer detection method based on local standard deviation and logarithmic computing. The method decreased the response values of comers near the candidate interested points through computing the logarithms of gradient, so it can detect comers with different types more effectively. And then, according to the statistical features of the standard deviation, it redefines the interest value function. The function could avoid subjectively selecting the value of parameters and it could directly judge whether a candidate interested point is a comer, which makes the algorithm more objective. The experimental results show that the method can effectively detect the comers with various types and it has a more accurate effect of positioning.