How to detect specific spots from image and crop in multiple images in python

Question

I am trying to detect some spots from the image and save it in multiple images after cropping.

Image:

I just want to crop wbc.

Script: I am trying but not getting the idea.

import cv2
import numpy as np;

# Read image
im = cv2.imread("C:/Users/Desktop/MedPrime_Tech_Work/tag-145218-Default-10X.jpg", cv2.IMREAD_GRAYSCALE)

# Set up the detector with default parameters.
detector = cv2.SimpleBlobDetector_create()

# Detect blobs.
keypoints = detector.detect(im)

print (keypoints)

# Draw detected blobs as red circles.
# cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS ensures the size of the circle corresponds to the size of blob
im_with_keypoints = cv2.drawKeypoints(im, keypoints, np.array([]), (0,0,255), cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)

# Show keypoints
cv2.imshow("Keypoints", im_with_keypoints)
cv2.waitKey(0)

My code is working but the problem is how to detect spots? as in shown in the image.

Thanks in advance. Please suggest something

Edit-1

Error Getting

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
<ipython-input-11-2754358a7c43> in <module>()
----> 1 import PyDIP as dip
      2 import PyDIP.PyDIPviewer as dv

ModuleNotFoundError: No module named 'PyDIP'

I am trying to install PyDIP but not unbale.

Answer 1

I don't have OpenCV here, but use PyDIP instead (I'm an author).

The detection in this case is fairly trivial because of the different color and sizes of the cells. Spaceman suggested to use the HSV color space. That is a good idea, but since this is so simple, I'm going to just use the individual green and blue channels instead. The "wbc" cells are very dark in the green channel, but not in the blue one. Everything that is black (outside of the field of view, and your drawings) is dark in both channels. So detecting the "wbc" and the "platelet" cells is a matter of finding dark regions in the green channel that are not dark in the blue. Next, a simple size criterion will exclude the "platelet" cells.

Finally, to crop, I group nearby detections (as these seem to belong together), and crop the groups from the image:

import PyDIP as dip

img = dip.ImageReadTIFF('/home/cris/tmp/cells')

# detect wbc
mask = dip.Erosion(img.TensorElement(2), dip.SE(7, 'elliptic'))
wbc = (img.TensorElement(1) < 50) & (mask > 50) # threshold green and blue channels, exact threshold values don't matter, color differences are obvious
wbc = dip.Closing(wbc, dip.SE(15, 'elliptic')) # fills small holes
wbc = dip.Opening(wbc, dip.SE(25, 'elliptic')) # removes small cells

# group and find bounding boxes
labs = dip.Label(dip.BinaryDilation(wbc, 2, 50)) # 50 is the half the distance between detections that belong together
labs *= wbc
m = dip.MeasurementTool.Measure(labs, features=['Minimum','Maximum'])

# crop
margin = 10 # space to add around detections when cropping
for obj in m.Objects():
    left = int(m[obj]['Minimum'][0]) - margin
    right = int(m[obj]['Maximum'][0]) + margin
    top = int(m[obj]['Minimum'][1]) - margin
    bottom = int(m[obj]['Maximum'][1]) + margin
    crop = img[left:right, top:bottom]
    dip.ImageWriteTIFF(crop, '/home/cris/tmp/cells%d'%obj)

This leads to the following small images:

Answer 2

When you say your code is working, I assume that means you're already detecting what you want to detect, and your question about cropping is just that you want to get images of the detected spots.

If I got that right, then remember that OpenCV images are just numpy arrays, so all you need is a subset of that array. Does this do what you want?

blob_images = list()
for kp in keypoints:
    x, y, r = *kp.pt, kp.size
    crop = im[y-r:y+r, x-r:x+r]
    blob_images.append(crop)

Now you should have a list of cropped images. From there, you can filter them so you only get white blood cells or save them with cv2.imwrite() or do whatever you want.

Be warned that crop is just a view into the im array, not a separate copy. This saves memory, but modifying one will modify the other. Use im[y-r:y+r, x-r:x+r].copy() if you need to decouple them.

Answer 3

I try to your able image to detect WBC which are in violet color.

import cv2
import numpy as np;

# Read image
im = cv2.imread("image.jpg")
hsv = cv2.cvtColor(im, cv2.COLOR_BGR2HSV)

I tried to convert it color space since i assume wbc's are always going to be in violet color. Since we have to detect object by color it better to convet it to other HSV color space you can read below link to know what HSV is below is output. cv2.imwrite("result1.jpg",hsv)

gray_image = cv2.cvtColor(hsv, cv2.COLOR_BGR2GRAY)
cv2.imwrite("result2.jpg",gray_image)

res,thresh_img=cv2.threshold(gray_image,210,255,cv2.THRESH_BINARY_INV) 

Here 210 is the threshold grayscale value for getting the spots which are white in color i.e WBC's

im2, contours, hierarchy = cv2.findContours(thresh_img,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)
cv2.imwrite("result3.jpg",thresh_img)

this image will all WBC's in black color. By using findContours we got the contours which are black in color.

count=0
original_image_copy=im.copy()    
for c in contours:
    area=cv2.contourArea(c)
    if area>10 and area<2500:
        (x,y),radius = cv2.minEnclosingCircle(c)
        center = (int(x),int(y))
        radius = int(radius)
        img = cv2.circle(im,center,radius,(0,255,0),10)
        cv2.circle(im,center,radius,(0,255,0),10)
        x, y, w, h = cv2.boundingRect(c)
        if w>20 and h>20:
            roi = original_image_copy[y:y+h, x:x+w]
            cv2.imwrite("images/roi"+str(count)+".jpg", roi) # make sure you have folder `images` in same folder in which you are running code.
        # cv2.drawContours(im, [c], 0,(0,255,0), 10)
cv2.imwrite("result4.jpg",im)

I know this wont be the perfect answer but you can tryto remove some noise in corner of the circle by clicking image in different way or you can use image processing function like morphology operattions like dilation.