如何检测全屏截图中是否存在指定图像？含需求与过往实现参考

Hey there! Let's build on the image detection logic you used for Diamond Dash—this problem is a perfect fit for template matching, which aligns exactly with what you were doing back then. Here's a practical, Python-based solution to detect that yellow success bar in your Facebook post screenshots and trigger the right Telegram notifications:

核心逻辑梳理 #

Your old script focused on "find single diamond → check surrounding → click". For this task, we can simplify it to "find fixed yellow bar → send corresponding Telegram alert"—since the bar hasn't changed in 5 years, it's a rock-solid success marker.

具体实现方案 #

I'll share two options based on your needs:

Option 1: PyAutoGUI (Quick & Simple, Fixed Resolution)

If your screenshots always have the same resolution, PyAutoGUI is super straightforward (you might already be familiar with it from the Diamond Dash days):

1. Prepare your template: Take a clear screenshot of the yellow success bar and save it as success_bar.png—make sure it matches the resolution of your post-screenshots.
2. Code implementation:

import pyautogui
import requests

# Fill in your Telegram details
TELEGRAM_BOT_TOKEN = "your_bot_token_here"
TELEGRAM_CHAT_ID = "your_chat_id_here"

def send_telegram_msg(text):
    url = f"https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendMessage"
    payload = {"chat_id": TELEGRAM_CHAT_ID, "text": text}
    requests.post(url, data=payload)

def send_telegram_screenshot(screenshot_path):
    url = f"https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendPhoto"
    with open(screenshot_path, "rb") as img_file:
        files = {"photo": img_file}
        payload = {"chat_id": TELEGRAM_CHAT_ID, "caption": "Post Status Screenshot"}
        requests.post(url, data=payload, files=files)

def check_post_success(screenshot_path, template_path):
    try:
        # Look for the template in the screenshot (confidence adjusts strictness)
        match_location = pyautogui.locateOnScreen(template_path, confidence=0.8)
        return match_location is not None
    except pyautogui.ImageNotFoundException:
        return False

# Main workflow
screenshot_file = "facebook_post_screenshot.png"  # Path to your full-screen screenshot
success_template = "success_bar.png"

if check_post_success(screenshot_file, success_template):
    send_telegram_msg("✅ Your content was published successfully!")
    send_telegram_screenshot(screenshot_file)
else:
    send_telegram_msg("❌ Content publish failed—please check the process!")
    send_telegram_screenshot(screenshot_file)

Install dependencies first: pip install pyautogui requests opencv-python (the confidence parameter needs OpenCV under the hood)

Option 2: OpenCV (More Accurate, Variable Resolution)

If your screenshots might have varying resolutions, OpenCV's multi-scale template matching handles this better:

import cv2
import numpy as np
import requests

# Fill in your Telegram details
TELEGRAM_BOT_TOKEN = "your_bot_token_here"
TELEGRAM_CHAT_ID = "your_chat_id_here"

def send_telegram_msg(text):
    url = f"https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendMessage"
    payload = {"chat_id": TELEGRAM_CHAT_ID, "text": text}
    requests.post(url, data=payload)

def send_telegram_screenshot(screenshot_path):
    url = f"https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendPhoto"
    with open(screenshot_path, "rb") as img_file:
        files = {"photo": img_file}
        payload = {"chat_id": TELEGRAM_CHAT_ID, "caption": "Post Status Screenshot"}
        requests.post(url, data=payload, files=files)

def check_success_with_opencv(screenshot_path, template_path):
    # Load images
    main_img = cv2.imread(screenshot_path)
    template_img = cv2.imread(template_path)
    template_h, template_w = template_img.shape[:2]

    # Search for template across different scales
    best_match = None
    for scale in np.linspace(0.8, 1.2, 20)[::-1]:
        resized_img = cv2.resize(main_img, (int(main_img.shape[1] * scale), int(main_img.shape[0] * scale)))
        scale_ratio = main_img.shape[1] / float(resized_img.shape[1])
        
        # Stop if resized image is smaller than template
        if resized_img.shape[0] < template_h or resized_img.shape[1] < template_w:
            break
        
        # Run template matching
        match_result = cv2.matchTemplate(resized_img, template_img, cv2.TM_CCOEFF_NORMED)
        _, max_match_val, _, max_match_loc = cv2.minMaxLoc(match_result)
        
        # Update best match if current is better
        if best_match is None or max_match_val > best_match[0]:
            best_match = (max_match_val, max_match_loc, scale_ratio)

    # Return True if match confidence is high enough
    return best_match is not None and best_match[0] >= 0.8

# Main workflow
screenshot_file = "facebook_post_screenshot.png"
success_template = "success_bar.png"

if check_success_with_opencv(screenshot_file, success_template):
    send_telegram_msg("✅ Your content was published successfully!")
    send_telegram_screenshot(screenshot_file)
else:
    send_telegram_msg("❌ Content publish failed—please check the process!")
    send_telegram_screenshot(screenshot_file)

Install dependencies: pip install opencv-python numpy requests

Key Tips for Reliability #

• Perfect your template: Capture the entire yellow bar without extra surrounding elements—this drastically reduces false matches.
• Tweak the confidence threshold: Adjust the 0.8 value in both scripts. Higher = stricter matching (less false positives), lower = more lenient (catches slightly distorted bars).
• Test with edge cases: Run a few test runs with failed post screenshots to make sure the script doesn't incorrectly flag success.

内容的提问来源于stack exchange，提问作者Geppettvs D'Constanzo