Edge linking and boundary detection

In order to extract meaningful information from images, image processing and computer vision require the critical processes of edge connecting and boundary detection. They entail converting unprocessed pixel data into more complex representations like contours, curves, and lines.
 

Edge Detection

Understanding edge detection is crucial before moving on to edge connecting. This method finds the pixels in the image where the intensity has changed significantly. Typical methods for detecting edges include:

Methods based on gradients: Prewitt, Sobel, and Canny operators
Gaussian Laplacian (LoG): detects blobs as well as edges.

Edge Linking

Edge linking creates object boundaries by joining fragmented edge pixels into continuous curves. The objective is to separate edge pixels from distinct objects while grouping those that belong to the same object. Typical methods include of:

• Local processing:
  1. Gradient direction: Connect edge pixels with similar gradient directions.
  2. Hysteresis thresholding: Use two thresholds to classify edges as strong or weak. Link weak edges to strong ones.
• Global processing:
  1. Hough transform: Detects lines and curves by accumulating votes in a parameter space.
  2. Active contours (snakes): Deformable curves that evolve to fit object boundaries.

Boundary Detection

The goal of boundary detection is to precisely identify the contours that divide different items in a picture. To get more exact bounds, edge linking findings are frequently refined. Methods consist of:
 

• Contour following is the process of tracking an object's perimeter by following its associated edge pixels.
• Region growth is the process of creating a region by repeatedly adding nearby pixels with comparable characteristics to a starting pixel.
• The watershed algorithm uses the topography of an intensity image to divide an image into areas.