Image Processing #12 - Detecting the center of text or curved line

 There are times when you need information about the center line of text or curved parts of an image.

I'm making a drawing robot these days. After the robot drew the picture, I needed this method to sign it. If the center line pixel location information of the signature image is grasped, the robot can naturally sign by drawing a pen while passing through these pixels.

Even when developing an autonomous driving robot, in order not to deviate from the road designated by the robot, it may be possible to identify the center line of the road and move it around this line.

Implementing this function in OpenCV is quite simple.

opencv-contrib-python not opencv-python

There is one thing to note. It is recommended to install opencv-contrib-python instead of opencv-python, which is normally installed to use OpenCV in Python. In this example, a function called cv2.ximgproc.thinning is used, but ximgproc is not supported by opencv-python. opencv-contrib-python is installed with additional modules along with opencv-python. This additional module contains ximgproc.

If you have already installed opencv-python using pip or pip3, uninstall it and install opencv-contrib-python.

Note: Do not install both versions at the same time.

pip3 uninstall opencv-python
pip3 install opencv-contrib-python

<test image>

Execute the following code.

import cv2
import sys

def detect(image):
    # some preprocessing
    thin = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    thin = cv2.blur(thin, (10, 10))
    _, thin = cv2.threshold(thin, 220, 255, 0)

    # thin image to find clear contours
    thin = cv2.ximgproc.thinning(thin, thinningType=cv2.ximgproc.THINNING_GUOHALL)
    cv2.imshow("thin", thin)
    # dind contours
    # cnts, _ = cv2.findContours(thin, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    cnts, _ = cv2.findContours(thin, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
    # cnts, _ = cv2.findContours(thin, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

 
    for cnt in cnts:
        cv2.drawContours(image, [cnt], 0, (255, 0, 0), 3)  # blue


    return image  


image = cv2.imread("C:\\lsh\\study\\opencv\\halo.jpg")
if image is None:
    print('No image')
    sys.exit()
output = detect(image)
cv2.imshow("CenterLine", output)
cv2.waitKey(0)
cv2.destroyAllWindows()

You will see the following window.

<result window>

If you want to check in cantour units, you can draw a rectangle surrounding the cantour. Just add the following 2 lines to the code that draws the center line.

    for cnt in cnts:
        cv2.drawContours(image, [cnt], 0, (255, 0, 0), 3)  # blue
        [x,y,w,h] = cv2.boundingRect(cnt)
        cv2.rectangle(image,(x,y),(x+w,y+h),(0,0,255),2)

You will see the following window.

<center line image with surrounding box>