Abstract: The main shortcoming of traditional Canny operator is that, its parameters need to be set by manual. To solve this problem, it introduces an adaptive multiscale Gauss filter by using intensity image’s mean square value histogram, which can remove noise and remain edge character. It self-adaptably generates thresholds by intensity image’s gradient character information, and detects edge automatically. Experiments show that, compared with the traditional Canny operator, this method can keep edge information well, and has better practicability.

Key words: edge-detection, Canny operator, multiscale, histogram

摘要： 针对传统Canny算子的参数需要人为设定的不足，提出利用灰度图象的均方差直方图实现自适应多尺度高斯滤波，实现在去噪的同时保留图象边缘特征，并通过灰度图象梯度特征信息，自适应生成动态阈值，对边缘进行自动检测。实验结果表明，与传统Canny算子相比，该方法有效保留边缘信息，更具实用性。

关键词: 边缘检测, Canny算子, 多尺度, 直方图