我应该使用什么样的参数来查找和裁剪图像中的对象？

Question

I am new to deep learning and try to implement a ML algorithm for image clustering. The problem is that I can't crop the objects in an image in Python using OpenCV. Here is the code I have implemented and it works for some objects if the color of the object is very different(in RGB values) from the background but it doesn't work for the image I need for ML algorithm. What kind of parameters should I have/change? Any suggestions?

import cv2
import numpy as np
from PIL import Image
import tkinter as tk
from tkinter import filedialog as fd
from tkinter import*
import random
#!/usr/bin/python
from PIL import Image
import sys


myFile = 'Path' + '/crop.png'
nr_of_im = 1
q = 0
r = 0
x_list = []
y_list = []
img = cv2.imread(myFile, cv2.IMREAD_UNCHANGED)

ret, thresh = cv2.threshold(cv2.cvtColor(img.copy(), cv2.COLOR_BGR2GRAY) , 30, 255, cv2.THRESH_BINARY)
contours, hier = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for contour in contours:
    print("len",len(contours))
    if cv2.contourArea(contour) > 80:
        x, y, w, h = cv2.boundingRect(contour)
        q = w
        r = h       
        x_list.append(x)
        y_list.append(y)
        font = cv2.FONT_HERSHEY_SIMPLEX
        ROI = img[y-10:y+10+h, x-10:x+10+w]
        ROI = cv2.resize(ROI,(300,300))
        file_all = "/images/%d.jpg"%nr_of_im
        nr_of_im += 1
        cv2.imwrite(file_all,ROI)

There are 21 objects in the image but the length of contours returns 1. The image looks like so

crop.png:

Answer 1

Your threshold is too low and produces a totally white image for me. You need to increase your threshold. Always view your thresholding to be sure it is working the way you expect. You can always remove the viewing later.

The following works for me using Otsu thresholding with a threshold value of 97. I get 21 contours.

Input:

import cv2
import numpy as np

# read image
img = cv2.imread('blocks.jpg')

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

# threshold
ret, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
print(ret)

# apply morphology fill and separate large regions and remove small ones
kernel = cv2.getStructuringElement(cv2.MORPH_RECT , (9,9))
morph = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT , (15,15))
morph = cv2.morphologyEx(morph, cv2.MORPH_OPEN, kernel)

# get contours
result = img.copy()
contours = cv2.findContours(morph, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]

# get count of contours
print(len(contours))

# draw bounding boxes on contours
for cntr in contours:
    x,y,w,h = cv2.boundingRect(cntr)
    cv2.rectangle(result, (x, y), (x+w, y+h), (0, 0, 255), 2)
    #print("x,y,w,h:",x,y,w,h)
    
# save results
cv2.imwrite("blocks_thresh.jpg", thresh)
cv2.imwrite("blocks_morphology.jpg", morph)
cv2.imwrite("blocks_bboxes.jpg", result)

# show thresh and result    
cv2.imshow("thresh", thresh)
cv2.imshow("morph", morph)
cv2.imshow("result", result)
cv2.waitKey(0)
cv2.destroyAllWindows()

Threshold image:

Morphology cleaned image:

Resulting bounding boxes from contours: