Cannyedge

Canny edge detector is a multi-stage edge detection operator that aims to identify sharp intensity changes in images while suppressing noise. Developed by John F. Canny in 1986, it has become a foundational method in computer vision for robust edge localization and extraction.

The standard implementation begins with smoothing the image with a Gaussian filter to reduce noise. It then computes the image gradient to measure edge strength and direction, typically via finite differences such as Sobel operators. The next stage applies non-maximum suppression to thin potential edges to a single-pixel width by keeping only local maxima along the gradient direction. After that, double thresholding classifies candidates into strong edges, weak edges, or none, based on two thresholds. Finally, edge tracking by hysteresis connects weak edges to strong ones if they are adjacent to a strong edge, producing the final edge map.

Parameters of the detector include the Gaussian kernel size or standard deviation and the two thresholds for hysteresis. While designed for grayscale images, it can be adapted to color images by processing luminance or converting to grayscale.

Canny edge detection is widely used in image analysis, feature extraction, object recognition, robotics, medical imaging, and computer vision preprocessing pipelines. Its advantages include good edge localization and low error rates in retaining true edges while suppressing noise; its limitations include sensitivity to threshold selection, potential poor performance on very low-contrast edges, and computational cost for high-resolution images. Numerous variants, including multi-scale and color-adapted versions, build on the original approach while trading off speed or robustness.