Removing circles, lines, and other patterns with OpenCV and Python, part 1 & 1/2

Why 1.5? Because I’m not addressing the issue of the overlay text and instead present a small improvement on removing the thick outlines.

import matplotlib.pyplot as plt
import numpy as np
import cv2
%matplotlib inline

# a little helper function to display our image in a bigger plot
def display_img(image):
    fig = plt.figure(figsize=(20,16))
    ax = fig.add_subplot(111)
    ax.imshow(image, cmap="gray")
img = cv2.imread('obfuscated.jpg', 0) #read in as a grayscale image
display_img(img)

png

Note that in the previous post I found out that a naive removal method and the Hough transform worked better than the Suzuki algorithm for contour location. The method I’m about to present should work regardless of the methods applied previously to the image, but it comes at a price. Let’s say I go with the naive method:

img_med = cv2.medianBlur(img, ksize=15)
ret, th = cv2.threshold(img_med, 80, 255, cv2.THRESH_BINARY)
display_img(th)

png

The secret here is to erode the image using a wide enough kernel (3x3 in our case). Normally erosion would make the outlines thinner, but because we’re working with a white background and black foreground it actually makes the outlines thicker. (Which is something we need since by first blurring and thresholding the image to extract the contours we have “watered down” the outlines significantly, leaving some of their traces behind after switching their pixels from black to white.)

kernel = np.ones((3,3),np.uint8)
eroded_img = cv2.erode(th,kernel,iterations=2)
display_img(eroded_img)

png

img_dest = img.copy()
img_dest[eroded_img == 0] = 255
display_img(img_dest)

png

And voilà! The geometric outlines are by and large gone. Unfortunately, this removes some of the letters as well (since they were covered by the outlines) but you can’t have everything ¯_(ツ)_/¯

Published 4 Apr 2019

Personal site; contains my incoherent ramblings, with trace amounts of code & dry jokes.
MK Mozgawa on Twitter