特征检测是图像处理和模式识别中非常重要的问题，其检测效果直接影响模式识别和分类。基于多尺度几何分析（MGA）的思想，提出了一种圆检测方法―圆特征域上奇异点算法。该算法首先将圆特征曲线变换到圆特征域上，然后在圆特征域上进行小波分析以找出奇异点，奇异点坐标即为待检圆的坐标。该方法克服了Hough变换对灰度图像圆检测需要考虑灰度阈值或梯度的限制，可直接对二值图像或灰度图像进行检测。最后分析、比较了该算法与Hough算法的不同。

Features detection is an important problem in image processing and pattern recognition, the detection result affects the pattern recognition and classification directly. Based on the idea of multi-scale geometric analysis, this paper introduces a new method to detect circles in digital images. Firstly, circles in images are transformed into circles feature domain. Then wavelet analysis is applied in the domain to detect point singularity. The coordinate of singular point is the coordinate of circle to be detected. Gray threshold or grads will be applied when Hough transform is used for gray image to detect circle. This disadvantage is overcome by the new method, and it is directly used for gray image or binary image to detect multi-circle. Finally, the difference between the new method and Hough transform is analyzed and compared.