Remove Black Background from Image with Grabcut - Python

Question

This is a follow on to a previously posted question about removing black backgrounds from .png images (Remove Background from Image - Python). The script I used in that Q&A works well but I still get minor inconsistencies in removing the black background exclusively while preserving everything inside the image contours. I tried a different script using the OpenCV grabcut algorithm, but I still get the same issue. I want to be able to remove the black background from the images without removing pixels inside image contours, here is the script I am using along with example input/output images upon running the script.

import numpy as np
import cv2
img = cv2.imread("C:\\users\\mdl518\\Desktop\\overhead_image.png") # Input 8-bit 3-channel image
      
mask = np.zeros(img.shape[:2], np.uint8) # img.shape[:2] = first, second values of img.shape
bgdModel = np.zeros((1,65), np.float64) # temporary array for background
fgdModel = np.zeros((1,65), np.float64) # temporary array for foreground
rect = (1,0,img.shape[1],img.shape[0]) # specify w,h beyond bounds of img.shape

# Grabcut algorithm to extract foreground (5 iterations) within region of interest
cv2.grabCut(img, mask, rect, bgdModel, fgdModel, 5, cv2.GC_INIT_WITH_RECT)

# Split RGB channels and create new alpha channel
r_channel, g_channel, b_channel = cv2.split(img) 
a_channel = np.where((mask==2)|(mask==0), 0, 255).astype('uint8')  
        
img_RGBA = cv2.merge((r_channel, g_channel, b_channel, a_channel)) # add alpha channel to image
cv2.imwrite(".//output_image.png", img_RGBA)  ## write new image to file

Input Image:

Output Image (using script above): Output Image (using "contour" method):

I am attempting to tweak my 'rect' bounds and the mask settings between 0 and 255, but I still cannot rectify this issue - The current settings seem to preserve the image the most without removing additional pixels, but it still does not preserve the whole image. Any assistance is most appreciated!

Answer 1

We can make a mask containing all of the not black in the image and use findContours to get the shapes of each black region. By selecting the largest contour we can mask out everything outside of the colored region.

import cv2
import numpy as np

# load image
img = cv2.imread("terrain.png");
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY);

# mask
mask = cv2.inRange(gray, 1, 255);

# get contours # OpenCV 3.4, in OpenCV 2* or 4* it returns (contours, _)
_, contours, _ = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE);

# get biggest contour
biggest = None;
biggest_size = -1;
for con in contours:
    area = cv2.contourArea(con);
    if area > biggest_size:
        biggest_size = area;
        biggest = con;

# make mask with biggest contour
new_mask = np.zeros_like(mask);
cv2.drawContours(new_mask, [biggest], -1, (255), -1);

# redraw with white
img[new_mask == 0] = (255,255,255);

# show image
cv2.imshow("Image", img);
cv2.imshow("Mask", mask);
cv2.imshow("New Mask", new_mask);
cv2.waitKey(0);