Skip to content
VisionKit
See detections

Hough shape & line detection

Finds the lines and circles hidden in an image.

Classic Hough transforms - no training, no models - pure C# on top of the OnlyCSharp educational library's own Canny/Sobel edge detectors.

Circle detection

A gradient-directed Hough circle vote finds round shapes and draws each one in green with a center cross. Below, VisionKit locates the coins in a synthetic fixture.

Input image of four coin-like discs on a flat background
1. Input
Four synthetic "coins".
The same image with detected circles outlined in green and center crosses
2. Detected circles
Green outlines + center crosses.

Line detection

Canny edges feed a Hough line vote; each detected straight line is drawn full-frame in red over the original. Here it recovers the walls, floor, eave, and roof edges of a simple house outline.

Input image of a house outline drawn with straight lines
1. Input
A house outline.
The same image with detected straight lines overlaid in red
2. Detected lines
Full-frame red Hough lines.

How it works

  • Edges first - the image is grayscaled and run through OnlyCSharp/1.8's real Sobel gradient and Canny edge detector.
  • Hough lines - every edge pixel votes for the family of lines through it in (rho, theta) space; peaks in the accumulator are the straight lines.
  • Hough circles - edge pixels vote along their gradient direction for candidate centers across a radius range; accumulator peaks become circles.

Command line

VisionKit is an offline console tool. The demos above were produced by:

visionkit circles coins-in.png coins-out.png --threshold 18 --low 30 --high 90 visionkit lines shapes-in.png shapes-out.png --threshold 80 --max 8