How to crop an image with upto 6 tissue samples using OpenCV?

I have an image from a public dataset (sample show below). I want to separate the two tissue samples using a tool to increase the number of instances in the dataset. Other images in the dataset have upto 6 tissues. Is there I can do that using OpenCV? Other tools are also welcomed.

click here to view the image

I tried this from one of the blogs I came across but it just creates a bounding box around the entire image. Maybe the thresholding scheme doesn't work well

import cv2

# read the input image
img = cv2.imread('pic.png')

# convert the image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# apply thresholding on the gray image to create a binary image
ret,thresh = cv2.threshold(gray,127,255,0)

# find the contours
contours, _ = cv2.findContours(gray,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)

# take the first contour
cnt = contours[0]

# compute the bounding rectangle of the contour
x,y,w,h = cv2.boundingRect(cnt)

# draw contour
gray = cv2.drawContours(gray,[cnt],0,(0,255,255),2)

# draw the bounding rectangle
gray = cv2.rectangle(gray,(x,y),(x+w,y+h),(0,255,0),2)

# display the image with bounding rectangle drawn on it
cv2.imshow("Bounding Rectangle", gray)
cv2.waitKey(0)
cv2.destroyAllWindows()

There are multiple issues:

• findContours expects white contours on black background - we have to invert polarity (and also select higher threshold):

   ret, thresh = cv2.threshold(gray, 230, 255, cv2.THRESH_BINARY_INV)
  

• contours, _ = cv2.findContours(gray is a bug - it should be contours, _ = cv2.findContours(thresh (find contours in thresh).

• Use cv2.RETR_EXTERNAL instead of cv2.RETR_TREE - we don't want to find nested contours, only the outer contour:

   contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  

• cnt = contours[0] return the first contour that may be tiny, we have to iterate all contours.

• We better filter contours by area - ignore small contours that are considered to be noise.

Code sample:

import cv2

# read the input image
img = cv2.imread('pic.png')

# convert the image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# apply thresholding on the gray image to create a binary image
#ret, thresh = cv2.threshold(gray, 127, 255, 0)
#ret, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_OTSU + cv2.THRESH_BINARY_INV)  # Invert polarity - findContours expects white contours
ret, thresh = cv2.threshold(gray, 230, 255, cv2.THRESH_BINARY_INV)  # Invert polarity - findContours expects white contours

# Consider adding morphological closing before cv2.findContours.

# find the contours
#contours, _ = cv2.findContours(gray, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)  # Why gray?
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)  # Use RETR_EXTERNAL instead of RETR_TREE - we don't want to find nested contours, only the outer contour

# take the first contour
#cnt = contours[0]  # Why the first? The fist may be a tiny contour

gray_as_bgr = cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)  # Prepare BGR image before drawing

# Iterate all contours
for cnt in contours:
    area = cv2.contourArea(cnt)  # Compute the area

    if area > 1000:  # Ignore contours with less than 1000 pixels - considered to be noise.
        # compute the bounding rectangle of the contour
        x, y, w, h = cv2.boundingRect(cnt)

        # draw contour        
        gray_as_bgr = cv2.drawContours(gray_as_bgr, [cnt], 0, (0, 255, 255), 2)

        # draw the bounding rectangle
        gray_as_bgr = cv2.rectangle(gray_as_bgr, (x, y), (x+w, y+h), (0, 255, 0), 2)  # Draw green rectangles

# display the image with bounding rectangle drawn on it
cv2.imshow("thresh", thresh)
cv2.imshow("Bounding Rectangle", gray_as_bgr)
cv2.waitKey(0)
cv2.destroyAllWindows()

Output gray_as_bgr: