如何使用OpenCV基于物体半径裁剪图像？Python新手的图像背景移除与裁剪技术问询

Hey there! Since you already know the basics of image preprocessing, let's tackle your two OpenCV questions clearly and with actionable code examples.

I'll walk through three methods, starting with the easiest for straightforward cases, then moving to more flexible options.

Simple Thresholding (Best for High-Contrast Images) #

If your foreground object stands out sharply against the background (like a dark object on a white background), thresholding works great:

import cv2
import numpy as np

# Load your image
img = cv2.imread('your_image.jpg')
# Convert to grayscale (required for thresholding)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Apply binary thresholding (adjust the 240 value to match your image)
# This turns pixels brighter than 240 white, and others black (inverted)
_, thresh = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY_INV)

# Clean up small noise with morphological operations (optional but helpful)
kernel = np.ones((3,3), np.uint8)
mask = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)

# Apply the mask to the original image to remove background
result = cv2.bitwise_and(img, img, mask=mask)

# If you want a transparent background (save as PNG)
b, g, r = cv2.split(result)
alpha_channel = mask  # Use our mask as the transparency layer
transparent_img = cv2.merge((b, g, r, alpha_channel))
cv2.imwrite('transparent_output.png', transparent_img)

Pro tip: Tweak the threshold value (240 in this case) until your object is fully captured in the thresholded image.

Contour-Based Removal (For Single, Distinct Objects) #

If your target is a single clear object, you can use contours to isolate it:

import cv2
import numpy as np

img = cv2.imread('your_image.jpg')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, thresh = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY_INV)

# Find all contours in the thresholded image
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# Pick the largest contour (assuming it's your target object)
largest_contour = max(contours, key=cv2.contourArea)

# Create a blank mask and draw the largest contour on it
mask = np.zeros_like(gray)
cv2.drawContours(mask, [largest_contour], 0, 255, -1)

# Apply the mask to get the object without background
result = cv2.bitwise_and(img, img, mask=mask)

# Show the result
cv2.imshow('Background Removed', result)
cv2.waitKey(0)
cv2.destroyAllWindows()

If you have multiple objects, add a filter to select contours by size or shape instead of just picking the largest.

Advanced: GrabCut Algorithm (For Complex Backgrounds) #

When the background and foreground blend together (like a person in a busy room), GrabCut is a powerful tool. It uses machine learning to separate foreground from background:

import cv2
import numpy as np

img = cv2.imread('your_image.jpg')
mask = np.zeros(img.shape[:2], np.uint8)

# Initialize background and foreground models (required for GrabCut)
bgd_model = np.zeros((1,65), np.float64)
fgd_model = np.zeros((1,65), np.float64)

# Define a rectangle around your object (adjust coordinates to fit)
# Format: (x, y, width, height)
object_rect = (50, 50, img.shape[1]-100, img.shape[0]-100)

# Run GrabCut (5 iterations is a good starting point)
cv2.grabCut(img, mask, object_rect, bgd_model, fgd_model, 5, cv2.GC_INIT_WITH_RECT)

# Refine the mask: mark sure/probable background as 0, foreground as 1
refined_mask = np.where((mask == 2) | (mask == 0), 0, 1).astype('uint8')
# Apply the refined mask to the image
result = img * refined_mask[:, :, np.newaxis]

cv2.imshow('GrabCut Result', result)
cv2.waitKey(0)
cv2.destroyAllWindows()

You can manually refine the mask by drawing on it if the initial result isn't perfect—look up cv2.GC_INIT_WITH_MASK for that step.

To crop around an object using its radius, first we need to find the object's center and radius, then crop the area around it. Here's how:

import cv2
import numpy as np

img = cv2.imread('your_image.jpg')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, thresh = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY_INV)

# Find contours and select the largest one
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
largest_contour = max(contours, key=cv2.contourArea)

# Get the minimum enclosing circle around the object
(center_x, center_y), radius = cv2.minEnclosingCircle(largest_contour)
center = (int(center_x), int(center_y))
radius = int(radius)

# Calculate crop boundaries (make sure we don't go outside the image)
x_start = max(0, int(center_x - radius))
y_start = max(0, int(center_y - radius))
x_end = min(img.shape[1], int(center_x + radius))
y_end = min(img.shape[0], int(center_y + radius))

# Crop the image to the square around the circle
cropped_img = img[y_start:y_end, x_start:x_end]

# Optional: Draw the circle on the original image to verify
cv2.circle(img, center, radius, (0, 255, 0), 2)
cv2.imshow('Object with Circle', img)
cv2.imshow('Cropped Image', cropped_img)
cv2.waitKey(0)
cv2.destroyAllWindows()

If you want a circular crop instead of a square, add a mask step:

# Create a circular mask for the cropped image
circular_mask = np.zeros(cropped_img.shape[:2], np.uint8)
crop_center = (int(radius), int(radius))  # Center of the cropped square
cv2.circle(circular_mask, crop_center, radius, 255, -1)

# Apply the mask to get a circular crop
circular_cropped = cv2.bitwise_and(cropped_img, cropped_img, mask=circular_mask)
cv2.imshow('Circular Crop', circular_cropped)
cv2.waitKey(0)

内容的提问来源于stack exchange，提问作者Sai Chand