如何从图像中删除最大轮廓标记的 object 并将其另存为单独的图像？

Question

Our objective is to classify plants in a field based on its leaves. We have trained our model on segmented images (these images only have a leaf and black background). But the live feed from the camera will look like this:

So our idea is to find the biggest contour, separate the leaf marked by it and give it a black background.

This is kinda what we are trying to achieve (except the small leaf popping in):

Our approach was to draw a bounding box around the leaf and form a new separate frame. This is our code:

def nothing(useless=None):
    pass

cv2.namedWindow("Mask")
cap = cv2.VideoCapture(0)

cv2.createTrackbar('R_l','Mask',26,255,nothing)
cv2.createTrackbar('G_l','Mask',46,255,nothing)
cv2.createTrackbar('B_l','Mask',68,255,nothing)

cv2.createTrackbar('R_h','Mask',108,255,nothing)
cv2.createTrackbar('G_h','Mask',138,255,nothing)
cv2.createTrackbar('B_h','Mask',155,255,nothing)

while True:

    R_l = cv2.getTrackbarPos('R_l', 'Mask')
    G_l = cv2.getTrackbarPos('G_l', 'Mask')
    B_l = cv2.getTrackbarPos('B_l', 'Mask')

    R_h = cv2.getTrackbarPos('R_h', 'Mask')
    G_h = cv2.getTrackbarPos('G_h', 'Mask')
    B_h = cv2.getTrackbarPos('B_h', 'Mask')

    _,frame = cap.read()
    blurred_frame = cv2.blur(frame,(5,5),0)

    hsv_frame = cv2.cvtColor(blurred_frame,cv2.COLOR_BGR2HSV)

    low_green = np.array([R_l, G_l, B_l])
    high_green = np.array([R_h, G_h, B_h])
    green_mask = cv2.inRange(hsv_frame, low_green, high_green)
    green = cv2.bitwise_and(frame, frame, mask=green_mask)

    contours,_ = cv2.findContours(green_mask,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)

    try:
        sorted_ = sorted(contours,key=cv2.contourArea,reverse=True)
        biggest = sorted_[0]
        cv2.drawContours(frame,biggest,-1,(255,0,0),1)
    except :
        pass
    #kernel = np.zeros(frame.shape(), np.uint8)
    x,y,w,h=cv2.boundingRect(biggest)
    roi= frame[y:y+h, x:x+w] 
    blurred_frame1 = cv2.blur(roi,(5,5),0)
    hsv_frame1 = cv2.cvtColor(blurred_frame1,cv2.COLOR_BGR2HSV)
    low_green1 = np.array([R_l, G_l, B_l])
    high_green1 = np.array([R_h, G_h, B_h])
    green_mask1 = cv2.inRange(hsv_frame1, low_green, high_green)
    green1= cv2.bitwise_and(roi,roi, mask=green_mask1)


    cv2.imshow("frame",frame)
    cv2.imshow("Mask",green1)
    key = cv2.waitKey(1)
    if key == 27:
        break
cap.release()
cv2.destroyAllWindows()

How can we prepare the desired image?

Answer 1

You're on the right track. I suggest using HSV color thresholding with a lower/upper threshold to isolate the green leaves. To determine the lower/upper HSV color threshold ranges, I used the HSV color thresholder script from a previous answer . This will give us a binary mask. From here we perform morphological operations to smooth the image and remove noise. Next we find contours and sort using contour area. We extract the largest contour, draw this onto a blank mask, then bitwise-and to get color. From here we find the bounding rectangle coordinates on the mask then crop the ROI from the color image using Numpy slicing. Here's the result

Code

import numpy as np
import cv2

# Read image, create blank masks, color threshold
image = cv2.imread('1.jpg')
blank_mask = np.zeros(image.shape, dtype=np.uint8)
original = image.copy()
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
lower = np.array([0, 18, 0])
upper = np.array([88, 255, 139])
mask = cv2.inRange(hsv, lower, upper)

# Perform morphological operations
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
opening = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel, iterations=1)
close = cv2.morphologyEx(opening, cv2.MORPH_CLOSE, kernel, iterations=1)

# Find contours and filter for largest contour
# Draw largest contour onto a blank mask then bitwise-and
cnts = cv2.findContours(close, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
cnts = sorted(cnts, key=cv2.contourArea, reverse=True)[0]
cv2.fillPoly(blank_mask, [cnts], (255,255,255))
blank_mask = cv2.cvtColor(blank_mask, cv2.COLOR_BGR2GRAY)
result = cv2.bitwise_and(original,original,mask=blank_mask)

# Crop ROI from result
x,y,w,h = cv2.boundingRect(blank_mask)
ROI = result[y:y+h, x:x+w]

cv2.imshow('result', result)
cv2.imshow('ROI', ROI)
cv2.waitKey()