图片中的笔画检测算法检测线条和曲线

Question

I am looking for an algorithm that detects all lines including curves etc. in a picture so i can redraw it using my software in a drawing program like paint. For now i only want it to be repainted in black and white. My approach was to make a stencil of the picture and try to read all black pixels as lines and finally paint it. The line calculation goes something like that:

 * for every pixel
 *      Point p = (x, y)
 *      List<Point> line
 *      while p is not marked
 *          mark p
 *          p = adjacent darkest pixel //brightness of a pixel is calculated by pixel luminance divided by 2 + luminance of the 8 adjacent pixels divided by 16
 *          add p to line
 *      end while
 *      draw line
 * end for

My approach works, but not very well. Some outlines get detected as two lines.

Do you have some improvements to my algorithm, or an even better one?

Answer 1

Try Canny Edge Detection which is a popular edge detection algorithm. It's already implemented in OpenCV as cv2.Canny() . Using a screenshoted input image, here's the result:

Input image

Result (inverted and non-inverted version)

Here's an implementation in Python OpenCV

import cv2

# Load image, convert to grayscale, and perform Canny edge detection
image = cv2.imread('1.png')
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
canny = 255 - cv2.Canny(gray, 120, 255, 1)

# Show image
cv2.imshow('canny', canny)
cv2.waitKey()

Note: To automatically determine the lower and upper thresholds, take a look at Zero-parameter, automatic Canny edge detection with Python and OpenCV