Masked out large irregular shape from image with Python-CodePudding

The objective is to remove large irregular area and maintained only character in the image.

For example, given the following

and the expected masked output

I have the impression this can be achieved as below

import cv2
import numpy as np
from matplotlib import pyplot as plt

dpath='remove_bg1.jpg'

img = cv2.imread(dpath)

img_fh=img.copy()
cv2.bitwise_not(img_fh,img_fh)
ksize=10
kernel = np.ones((ksize,ksize),np.uint8)
erosion = cv2.erode(img_fh,kernel,iterations = 3)
invertx = cv2.bitwise_not(erosion)
masked = cv2.bitwise_not(cv2.bitwise_and(img_fh,invertx))

all_image=[img,invertx,masked]
ncol=len(all_image)
for idx, i in enumerate(all_image):
    plt.subplot(int(f'1{ncol}{idx 1}')),plt.imshow(i)

plt.show()

which produce

Clearly, the code above did not produced the expected result.

May I know how to address this issue properly?

CodePudding user response：

To remove the unwanted blob, we must create a mask such that it encloses it completely.

Flow:

Inversely binarize the image (such that you have a white foreground against dark background)
Dilate the image (since the blob makes contact with letter 'A', it has to be isolated )
Find contour with the largest area
Draw the contour on an another 1-channel image and thicken it (dilation)
Pixel Assignment: Pixels containing the dilated blob are made white on the original image

Code:

im = cv2.imread('stained_text.jpg')
im2 = im.copy()
gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)

# inverse binaraization
th = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV   cv2.THRESH_OTSU)[1]

Notice the blob region touching the letter 'A'. Hence to isolate it we perform erosion using an elliptical kernel

# erosion
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5,5))
erode = cv2.erode(th, kernel, iterations=2)

# find contours
contours, hierarchy = cv2.findContours(erode, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
# Contour of maximum area
c = max(contours, key = cv2.contourArea)

# create 1-channel image in black 
black = np.zeros((im.shape[0], im.shape[1]), np.uint8)
# draw the contour on it
black = cv2.drawContours(black, [c], 0, 255, -1)
# perform dilation to have clean border
# we are using the same kernel
dilate = cv2.dilate(black, kernel, iterations = 3)

# assign the dilated area in white over the original image
im2[dilate == 255] = (255,255,255)

This was just one of the many possible ways on how to proceed. The key thing to note is how to isolate the blob.