从经典的Harris角点检测方法不难看出，该算法的稳定性和k有关，而k是个经验值，不好把握，浮动也有可能较大。鉴于此，改进的Harris方法（）直接计算出两个特征值，通过比较两个特征值直接分类，这样就不用计算Harris响应函数了。

另一方面，我们不再用非极大值抑制了，而选取容忍距离：容忍距离内只有一个特征点。该算法首先选取一个具有最大最小特征值的点（即：max（min（e1，e2）），e1，e2是harris矩阵的特征值）作为角点，然后依次按照最大最小特征值顺序寻找余下的角点，当然和前一角点距离在容忍距离内的新角点呗忽略。-Difficult to see　the classic Harris corner detection method, the algorithm stability and k, whereas k is an experience, not good grasp, floating there may be larger. In view of this, an improved method of Harris () directly calculates two characteristic values, by comparing two eigenvalues classification, so do not calculate the response function of Harris. On the other hand, we are no longer inhibited by non-maximum value,　the distance tolerance: tolerance distance is only one feature point. Firstly,　a point maximum and minimum eigenvalues have (ie: max (min (e1, e2)), e1, e2 is the characteristic value harris matrix) as the corners, followed in accordance with the maximum and minimum eigenvalues in order to find the remaining corner, of course, and the front corner of the distance in angular distance tolerate new point chanting ignored.