Detect textured background for contrasted images

Question

I have two types of images, and they were both contrasted.

Type 1: (white and clean background)

Type 2: (some gray texture in background)

I could apply Gaussian blur and Thresholding to process the type 2 image to adjust it to the white background like type 1 as follow code:

type2_img = cv2.imread(type2.png)
# convert to grayscale
gray = cv2.cvtColor(type2_img , cv2.COLOR_BGR2GRAY)
# threshold the image using Otsu's thresholding
thresh = cv2.threshold(gray.copy(), 0, 255, cv2.THRESH_OTSU)[1]
# convert back to RGB
type2_img = cv2.cvtColor(thresh,cv2.COLOR_GRAY2RGB)

And, I could get the following result :

However, I do not wish to apply the same method for type 1 images, since it is already satisfied condition.

So, are there any image processing methods within OpenCV that can differentiate type 2 images from type 1?

Answer 1

The type 1 image you provided is grayscale and not completely black & white image. Yet the last threshold output image is black and white.

Based on your question my understanding is both of these should as classified as type 1 and the type 2 image example should classified as type 2.

If it was the case that type 1 was always black and white you could count the number of 0's and 1's in image and check if their sum equaled to total pixel count of image.

An option here is to classify two types of image without modifying images is to use the percentage of black and white pixels of image histogram.

Code

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

#img_path = r'hist_imgs/1.png'
#img_path = r'hist_imgs/2.png'
img_path = r'hist_imgs/3.png'

img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)
cv2.imshow('Image', img)

img_pixel_count = img.shape[0] * img.shape[1]
print(img.shape)
print(img_pixel_count)

h = np.array(exposure.histogram(img, nbins=256))

bw_count = h[0][0] + h[0][255]
other_count = img_pixel_count - bw_count

print('BW PIXEL COUNT: ',  bw_count)
print('OTHER PIXEL COUNT: ',  other_count)

bw_percentage = (bw_count * 100.0) / img_pixel_count
other_percentage = (other_count * 100.0) / img_pixel_count

print('BW PIXEL PERCENTAGE: ',  bw_percentage)
print('OTHER PIXEL PERCENTAGE: ',  other_percentage)

differentiate_threshold = 3.0
if other_percentage > differentiate_threshold:
    print('TYPE 2: GRAYSCALE')
else:
    print('TYPE 1: BLACK AND WHITE')

plt.hist(img.ravel(), 256, [0, 256])
plt.title('Histogram')
plt.show()

Input Images

Image 1:

Image 2:

Image 3:

Code Output

Image 1:

(154, 74)
11396
BW PIXEL COUNT:  11079
OTHER PIXEL COUNT:  317
BW PIXEL PERCENTAGE:  97.21832221832221
OTHER PIXEL PERCENTAGE:  2.781677781677782
TYPE 1: BLACK AND WHITE

Image 2:

(38, 79)
3002
BW PIXEL COUNT:  1543
OTHER PIXEL COUNT:  1459
BW PIXEL PERCENTAGE:  51.39906728847435
OTHER PIXEL PERCENTAGE:  48.60093271152565
TYPE 2: GRAYSCALE

Image 3:

(38, 79)
3002
BW PIXEL COUNT:  3002
OTHER PIXEL COUNT:  0
BW PIXEL PERCENTAGE:  100.0
OTHER PIXEL PERCENTAGE:  0.0
TYPE 1: BLACK AND WHITE

Answer 2

To know if an image is blurried I would use the variance of the Laplacian.

import cv2

im1 = cv2.imread("s1.png")
im2 = cv2.imread("s2.png")

print(f"Im1: {cv2.Laplacian(im1, cv2.CV_64F).var()}")
print(f"Im2: {cv2.Laplacian(im2, cv2.CV_64F).var()}")

Output:

Im1: 10493.1934011934
Im2: 17803.672073381236

I would set a threshold in the middle of each class looking through some train set or examples.

Hope it works!