Template matching in Computer Vision - ML Experiment: Train & Evaluate

Experiment - Template matching

Problem:You want to find a small image (template) inside a bigger image using template matching.

Current Metrics:The current method finds the template but often gives false matches or misses the correct location.

Issue:The template matching is not accurate enough, leading to wrong detections or no detection.

Your Task

Improve the template matching accuracy to correctly locate the template in the image with fewer false matches.

You must use OpenCV's template matching functions.

You cannot change the template or the main image.

You can only adjust the matching method and threshold.

Hint 1

Hint 2

Hint 3

Solution

Computer Vision

import cv2
import numpy as np

# Load the main image and template in grayscale
image = cv2.imread('main_image.jpg', cv2.IMREAD_GRAYSCALE)
template = cv2.imread('template.jpg', cv2.IMREAD_GRAYSCALE)

# Get template width and height
w, h = template.shape[::-1]

# Apply template matching using normalized correlation coefficient
result = cv2.matchTemplate(image, template, cv2.TM_CCOEFF_NORMED)

# Set threshold for detecting good matches
threshold = 0.8

# Find locations where matching result is above threshold
locations = np.where(result >= threshold)

# Draw rectangles on matches
for pt in zip(*locations[::-1]):
    cv2.rectangle(image, pt, (pt[0] + w, pt[1] + h), (255, 255, 255), 2)

# Save or show the result image
cv2.imwrite('matched_result.jpg', image)

# Calculate number of matches found
num_matches = len(locations[0])
print(f'Number of matches found: {num_matches}')

Switched to TM_CCOEFF_NORMED matching method for better accuracy.

Added a threshold of 0.8 to filter out weak matches.

Visualized matches by drawing rectangles on the original image.

Results Interpretation

Before: Template matching gave many false matches or missed the template location.

After: Using TM_CCOEFF_NORMED and thresholding, the model found 3 correct matches with fewer false positives.

Choosing the right matching method and applying a threshold helps improve template matching accuracy by reducing false detections.

Bonus Experiment

Try using multi-scale template matching to find the template even if it appears at different sizes in the image.

💡 Hint

Resize the template to different scales and apply template matching at each scale, then pick the best match.