Edge detection is a crucial approach in image processing, machine vision, and computer vision.

Edge detection refers to a set of mathematical techniques for detecting edges, or curves in a digital picture when the brightness of the image abruptly changes or, more formally, has discontinuities. Step detection is the issue of identifying discontinuities in one-dimensional signals, while change detection is the problem of finding signal discontinuities across time. In image processing, machine vision, and computer vision, edge detection is a critical technique, especially in the fields of feature identification and extraction.

The goal of detecting sharp changes in picture brightness is to record significant events and changes in the world's characteristics. Discontinuities in picture brightness are expected to correlate to discontinuities in-depth, discontinuities in surface orientation, changes in material characteristics, and fluctuations in scene light given relatively generic assumptions for an image generation model.

In an ideal world, applying an edge detector to an image would result in a collection of linked curves that indicate object borders, surface marking boundaries, and curves that correspond to surface orientation discontinuities. Applying an edge detection method to a picture can minimize the quantity of data that has to be processed and therefore filter out information that isn't as vital while retaining the image's crucial structural features. If the edge detection stage is successful, the job of understanding the information contained in the original image may be significantly streamlined. However, such perfect edges are not always possible to get from real-life pictures of modest complexity.

Edges recovered from non-trivial pictures are frequently impeded by fragmentation, which results in unconnected edge curves, missing edge segments, and false edges that do not correlate to important events in the image, complicating the process of understanding the image data. One of the most basic processes in image processing, image analysis, picture pattern recognition, and computer vision approaches is edge detection.

Viewpoint-dependent or viewpoint-independent edges can be retrieved from a two-dimensional picture of a three-dimensional scene. The intrinsic features of three-dimensional objects, such as surface marks and form, are generally reflected by a perspective-independent edge. The geometry of the scene, such as objects occluding one another, is generally reflected by a perspective-dependent edge, which varies as the viewpoint changes.

The border between a block of red and a block of yellow, for example, is a typical edge. A line, on the other hand, can be a tiny number of pixels of a variable hue on an otherwise constant backdrop (as can be retrieved by a ridge detector). As a result, there may be one edge on either side of a line in most cases.

Edge detection may be done in a variety of ways, with Prewitt edge detection, Sobel edge detection, Laplacian edge detection, and Canny edge detection being some of the most popular.

Prewitt Edge Detection

This is a popular edge detector that is used to identify horizontal and vertical edges in pictures.

Sobel Edge Detection

This makes use of a filter that emphasizes the filter's center. It is one of the most often used edge detectors, and it reduces noise while also providing distinguishing and edge response.

Laplacian Edge Detection

The Laplacian edge detectors are different from the edge detectors previously mentioned. Only one filter is used in this technique (also called a kernel). Laplacian edge detection executes second-order derivatives in a single pass, making it susceptible to noise. Before using this approach, the picture is smoothed with Gaussian smoothing to avoid this susceptibility to noise.

Canny Edge Detection

This is the most widely utilized, highly successful, and complicated approach in comparison to many others. It's a multi-stage method for detecting and identifying a variety of edges. The steps of the Canny edge detection method are shown below. It transforms the picture to grayscale, eliminates noise (since edge detection using derivatives is susceptible to noise), calculates the gradient (which aids in identifying the edge strength and direction), and last, turns the image to grayscale. It employs non-maximum suppression to narrow the image's edges, a double threshold to detect the image's strong, weak, and irrelevant pixels, and hysteresis edge tracking to help transform weak pixels into strong pixels only if they are surrounded by strong pixels.