You clearly have a low-contrast image with all brightnesses in the range around 64-88, i.e. 25% to 33% of the 0..255 range. So my first instinct is to stretch the contrast so that 64 becomes zero and 88 becomes 255. I am doing it with ImageMagick here because it is conceptual and you/we can do it with OpenCV later if necessary.
magick RPU7I.png -level 25,33% 1.jpg
We could then copy that image, blur it and subtract to remove the low-frequency background features. Firstly with a small blur:
magick 1.jpg \( clone -blur 0x10 \) -compose difference -composite -auto-level 2.jpg
Then with a larger blur:
magick 1.jpg \( clone -blur 0x80 \) -compose difference -composite -auto-level 3.jpg
Or, we could go back to our original contrast-stretched image 1.jpg and try CLAHE on that:
magick 1.jpg -clahe 20x20% 128 20 result.jpg
Any image-processing wizards welcome to copy, steal, disagree, improve, extend, edit any concepts here...
CodePudding user response：
Here is another approach using division normalization followed by some dynamic range stretching in Python/OpenCV.
- Read the input
- Convert to grayscale
- Blur with relatively large sigma
- Divide the blurred image by the input
- Stretch the dynamic range
- Save the result
Input:
import cv2
import numpy as np
import skimage.exposure
# load image
img = cv2.imread("monocytes.png")
# convert to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# blur
blur = cv2.GaussianBlur(gray, (0,0), sigmaX=39, sigmaY=39)
# divide
divide = cv2.divide(blur, gray, scale=255)
divide = 255 - divide
# stretch
maxval = np.amax(divide)/4
stretch = skimage.exposure.rescale_intensity(divide, in_range=(0,maxval), out_range=(0,255)).astype(np.uint8)
# write result to disk
cv2.imwrite("monocytes_division.png", divide)
cv2.imwrite("monocytes_stretch.jpg", stretch)
# display it
cv2.imshow("gray", gray)
cv2.imshow("divide", divide)
cv2.imshow("stretch", stretch)
cv2.waitKey(0)
cv2.destroyAllWindows()
Stretched Result:
