Abstract: Improving the precision of corner detection can get higher precision of image registration in image mosaics.An improved Harris-based corner detection algorithm is presented in the paper.It makes curve fitting to the RHarris values of all pixels in every line on x-direction and every column on y-direction of the image respectively based on the Harris algorithm.And then，the curve is analyzed.When the RHarris value of a certain pixel is at the wave peak position on both x and y direction，it is detected as a corner.The experiments show that this method can avoid corners redundancies or loss brought by uncorrected threshold selection effectively and improve the precision of corners detection.

Key words: Harris operator, corner detection, curve fitting, degree of freedom

摘要： 图像拼接中，提高角点检测的精确度可以提高配准的精度。在Harris算法的基础上，提出了一种改进的角点检测算法，算法首先分别对图像中每行、每列上所有像素点的Rharris进行X、Y方向的曲线拟合，然后对拟合后的曲线进行分析，若某个像素点的Rharris值在两个方向上都处于“波峰”位置，则将该像素点检测为角点。实验结果表明该算法可以避免阈值的选择，有效地克服了阈值选择不当造成的角点冗余或丢失，提高了角点检测的精度。

关键词: Harris算子, 角点检测, 曲线拟合, 自由度