Feature Detection, Part 2: Laplace & Gaussian Operators

Edge detection in computer vision can be performed by analyzing the second derivative of image intensity, where edges correspond to zero-crossings. The Laplacian operator is introduced as a method for calculating this second derivative, and its discrete approximation is derived into a convolutional kernel. A significant challenge with this approach is its high sensitivity to image noise, which can create false positives. To mitigate this, a Gaussian filter is first applied to smooth the image and suppress noise before applying the Laplacian. This combined technique, known as the Laplacian of Gaussian (LoG), provides more robust and reliable edge detection.