Canny edge detection

woman

In this project I implemented the famous Canny edge detection algorithm. Following steps were involved:

Smoothing the image with a Gaussian filter to remove speckles and avoid spurious edges
Finding gradient magnitude and orientation at every pixel using convolution with a differentiation kernel
Deciding the low and high Canny algorithm thresholds based on image statistics (gradient magnitude mean and standard deviation)
Removing pixels with magnitude lower than lower threshold and non-maximum suppression to make all edges 1 pixel thick
Edge linking by starting at pixels with gradient magnitude higher than higher threshold and stopping at a pixel with gradient magnitude lower than lower threshold

Here are some results:

Wiki