Search code examples
pythonpython-3.xopencv3.0opencv

Crop colour image according to OTSU threshold

I have a colour image which I have sucessfully applied the OTSU thresholding method on its greyscale form to obtain the outline of the biscuit:

Original Colour image:

Image

OTSU Thresholded image:

OTSU

What I would like to do is extract from the colour image only the pixels within the black portion of the OTSU thresholded image, and save that as a new picture. So far, I have tried to extract using the BITWISE_NOT method and using 'thresh_inv' as the mask, however that only results in this image (greyscale + extra black background). I have also tried using the cannny contours method to identify the rough outline of the biscuit circle, and then drawing the contours over a blank image to hopefully try and overlay that over the original colour picture. This has also not worked.

Attempt_1

Here is my code so far, I would greatly appreciate any help as I've been trying to figure it out for ages.

import numpy as np
import cv2 as cv2

picture = cv2.imread('Cropped_6.png',0)
blurred_picture = cv2.GaussianBlur(picture, (15,15), cv2.BORDER_DEFAULT)
# canny_picture = cv2.Canny(blurred_picture, 41,41)
# cv2.imshow('Blurred', canny_picture)

# Simple Thresholding
threshold, thresh = cv2.threshold(blurred_picture, 105, 255, cv2.THRESH_OTSU)
# contours, hierarchies = cv2.findContours(thresh, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
thresh_inv = cv2.bitwise_not(thresh)
foreground = cv2.bitwise_and(picture, picture, mask=thresh_inv)

cv2.imwrite('Attempt_1.png',foreground)

# cv2.drawContours(blank_picture, contours[:0], -1, (0,0,255), 1)
cv2.imshow('Original', picture)
cv2.waitKey(0)

This is an example of what I'd end up with as a result (just the biscuit with a transparent background in the rectangle containing it, since afaik even circles are stored as rectangular images):

Desired result:

Ideal Result

Solution

  • 1st load the image as rgb. Then convert it into binary. since it has 3 channel then you need to stack your mask to form a array with 3 channel Then perform binary and operation on them.

    Try:

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

picture_rgb = cv2.imread('cropped6.png',1)
picture = cv2.cvtColor(picture_rgb, cv2.COLOR_BGR2GRAY)

blurred_picture = cv2.GaussianBlur(picture, (15,15), cv2.BORDER_DEFAULT)
# canny_picture = cv2.Canny(blurred_picture, 41,41)
# cv2.imshow('Blurred', canny_picture)

# Simple Thresholding
threshold, thresh = cv2.threshold(blurred_picture, 105, 255, cv2.THRESH_OTSU)
# contours, hierarchies = cv2.findContours(thresh, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
thresh_inv = cv2.bitwise_not(thresh)
stacked = np.dstack((thresh_inv,thresh_inv,thresh_inv))
img = cv2.bitwise_and(picture_rgb, stacked)
foreground = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
foreground[np.all(foreground == (0, 0, 0), axis=-1)] = (255,255,255)

cv2.imwrite('Attempt_1.png',foreground)

# # cv2.drawContours(blank_picture, contours[:0], -1, (0,0,255), 1)
# cv2.imshow('Original', picture)
# cv2.waitKey(0) 
plt.imshow(foreground, cmap='gray')

    To transparent it: (source: how to make white pixels transparent)

    from PIL import Image

img = Image.fromarray(np.uint8(foreground)).convert('RGBA')
img = img.convert("RGBA")
datas = img.getdata()

newData = []
for item in datas:
    if item[0] == 255 and item[1] == 255 and item[2] == 255:
        newData.append((255, 255, 255, 0))
    else:
        newData.append(item)

img.putdata(newData)
img.save("img2.png", "PNG")