Detect filled in black rectangles on patterned background with Python OpenCV

I'm trying to detect the location of these filled-in black rectangles using OpenCV.

Black rectangles on paper

I have tried to find the contours of these, but I think the background lines are also detected as objects. Also the rectangles aren't fully seperated (sometimes they touch a corner), and then they are detected as one, but I want the location of each of them seperately.

Here are the results I got, from the following code.

import numpy as np 
import cv2

image = cv2.imread("page.jpg")

result = image.copy()
gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

thresh = cv2.adaptiveThreshold(gray,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV,51,9)

cnts = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
for c in cnts:
    cv2.drawContours(thresh, [c], -1, (255,255,255), -1)

kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=4)

cnts = cv2.findContours(opening, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
for c in cnts:
    x,y,w,h = cv2.boundingRect(c)
    cv2.rectangle(image, (x, y), (x + w, y + h), (36,255,12), 3)

cv2.imshow('thresh', thresh)
cv2.imshow('opening', opening)
cv2.imshow('image', image)
cv2.waitKey()

Threshold

Opening

Result

As you can see, in the Opening image, the white rectangles are joined with the black ones, but I want them seperately. Then in the Result, it just detects a contour around the entire page.

As I said in the comments, no need for adaptive thresholding. Simple problems are best solved with simple solutions. A simple threshold would suffice in your case, but I guess you did not want to do that because of the lines? Is there any reason why you went with adaptive thresholding?

Here's my appraoch:

1. Read image
2. Define threshold
3. Generate an erosion rectangle with a width which you will need later
4. Erode the binary mask
5. Get contours
6. For every contour, correct the properties of the rectangle, to account for the erosion step (more on this in the next two pictures)
7. Print out, draw, save, go crazy...

Step number 6 is important, here's why: the erosion takes a chunk of your area, width and height. This can be clearly seen in the already accepted answer.

Before the correction, the rectangles will look like this:

Notice what I mentioned earlier, bad dimensions. By taking into account what has been eroded, we can closely estimate the actual rectangle:

I hope this helps you further, here's the code as a dump:

import cv2
%matplotlib qt
import matplotlib.pyplot as plt
import numpy as np
im = cv2.imread("stack.png") # read as BGR
imGray = cv2.imread("stack.png", cv2.IMREAD_GRAYSCALE) # read as gray
kernelSize = 20 # define the size of the erosion rectangle
smallRectangle = np.ones((kernelSize, kernelSize), dtype=np.uint8) # define the small erosion rectangle
mask = (imGray<130).astype("uint8") # get the mask based on a threshold, I think OTSU might work here as well
eroded = cv2.erode(mask, smallRectangle) # erode the image
contours, _ = cv2.findContours(eroded, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_NONE) # find contours
for i, cnt in enumerate(contours): # for every cnt in cnts
    # parse the rectangle parameters
    x,y,w,h = cv2.boundingRect(cnt)
    # correct the identified boxes to account for erosion
    x -= kernelSize//2
    y -= kernelSize//2
    w += kernelSize
    h += kernelSize
    box = (x,y,w,h) # assemble box back
    # draw rectangle
    im = cv2.rectangle(im, box, (0,0,255), 3)
    # print out results
    print(f"Rectangle {i}:\n center at ({x+w//2}, {y+h//2})\n width {w} px\n height {h} px")

Results of the printing:

Rectangle 0:
 center at (87, 471)
 width 42 px
 height 74 px
Rectangle 1:
 center at (158, 403)
 width 38 px
 height 75 px
Rectangle 2:
 center at (229, 403)
 width 45 px
 height 78 px
Rectangle 3:
 center at (121, 333)
 width 41 px
 height 77 px
Rectangle 4:
 center at (47, 259)
 width 43 px
 height 75 px
Rectangle 5:
 center at (82, 191)
 width 46 px
 height 77 px