Python中如何将图像叠加到背景图的指定四边形坐标位置？

Hey there! Great question—since you're working with an arbitrary quadrilateral (not a parallelogram), the Affine transform won't cut it here. Affine transforms only preserve parallel lines and can only map 3 points to 3 points, which limits you to shapes like rectangles or sheared parallelograms. For a 4-point quadrilateral, you need a Perspective Transform (also called a Homography), which is exactly designed for warping images to fit any four-cornered shape.

Here's a step-by-step implementation using PIL and NumPy (plus OpenCV for easy homography calculation—don't worry, it's straightforward):

Step 1: Import Required Libraries #

from PIL import Image
import numpy as np
# We'll use OpenCV to calculate the perspective transform matrix easily
import cv2

Step 2: Load Your Images #

# Replace with your actual image paths
background_img = Image.open("background.jpg")
foreground_img = Image.open("foreground.png")

Step 3: Define Your Coordinates #

Use the quadrilateral coordinates you provided, and map them to the four corners of your foreground image:

# Target quadrilateral on the background (your coordinates)
target_quad = np.array([
    [37., 119.],   # TOP LEFT
    [615., 65.],   # TOP RIGHT
    [648., 416.],  # BOTTOM RIGHT
    [70., 470.]    # BOTTOM LEFT
], dtype=np.float32)

# Source points: the four corners of your foreground image
fg_width, fg_height = foreground_img.size
source_points = np.array([
    [0, 0],        # TOP LEFT
    [fg_width, 0], # TOP RIGHT
    [fg_width, fg_height], # BOTTOM RIGHT
    [0, fg_height] # BOTTOM LEFT
], dtype=np.float32)

Step 4: Calculate the Perspective Transform & Warp the Foreground #

# Compute the homography matrix (maps source points to target quadrilateral)
homography_matrix, _ = cv2.findHomography(source_points, target_quad)

# Convert the foreground image to a NumPy array for OpenCV processing
foreground_np = np.array(foreground_img)

# Warp the foreground to fit the target quadrilateral
# The output size matches the background image dimensions
warped_foreground_np = cv2.warpPerspective(
    foreground_np,
    homography_matrix,
    (background_img.width, background_img.height)
)

# Convert the warped image back to a PIL Image
warped_foreground = Image.fromarray(warped_foreground_np)

Step 5: Composite the Warped Image onto the Background #

If your foreground has a transparent alpha channel, the paste method will respect it automatically when you pass the image as the mask:

# Paste the warped foreground onto the background
background_img.paste(warped_foreground, (0, 0), warped_foreground)

# Save or display the final result
background_img.save("final_result.jpg")
background_img.show()

Key Notes: #

• Coordinate Order: Make sure your source and target points follow the same sequence (e.g., top-left → top-right → bottom-right → bottom-left). Mismatched order will cause distorted warping.
• No OpenCV?: If you don't want to use OpenCV, you can manually calculate the perspective transform matrix using linear algebra, but it's far more code-heavy. OpenCV's findHomography is the simplest way to get the correct matrix.
• Transparency: If your foreground doesn't have an alpha channel, you might need to add one or create a custom mask to blend it smoothly with the background.

内容的提问来源于stack exchange，提问作者Morgan Roberts