Template matching in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - Template matching

Which metric matters for Template Matching and WHY

Template matching finds a small image inside a bigger one. The key metric is matching accuracy, which tells how often the template is correctly found. We also use precision and recall to understand if the method finds the right spots without many mistakes or misses.

Precision matters because we want to avoid false matches (wrong spots). Recall matters because we want to find all real matches. Balancing both helps us trust the results.

Confusion Matrix for Template Matching

      | Predicted Match    | Predicted No Match |
      |--------------------|--------------------|
      | True Positive (TP)  | False Negative (FN) |
      | False Positive (FP) | True Negative (TN)  |

      Example:
      Suppose we have 100 places where the template could appear.
      - TP = 70 (correctly found matches)
      - FP = 10 (wrong matches found)
      - FN = 20 (missed matches)
      - TN = 0 (not usually counted in template matching)

      Total samples = TP + FP + FN = 100
    

Precision vs Recall Tradeoff in Template Matching

If we set a strict matching threshold, we get high precision (few false matches) but low recall (miss many real matches).

If we set a loose threshold, recall improves (find more matches) but precision drops (more false matches).

Example: In quality control, missing a defect (low recall) is worse than a few false alarms (lower precision). So recall is more important.

In other cases, like face detection, false matches confuse the system, so precision is more important.

Good vs Bad Metric Values for Template Matching

• Good: Precision > 0.9 and Recall > 0.85 means most matches are correct and most real matches are found.
• Bad: Precision < 0.5 means many false matches, Recall < 0.5 means many missed matches.
• Accuracy alone is less useful because many non-match areas exist, inflating accuracy.

Common Pitfalls in Template Matching Metrics

• Accuracy paradox: High accuracy can happen if most image areas are non-matches, hiding poor matching performance.
• Data leakage: Testing on images too similar to training templates inflates metrics.
• Overfitting: Template matching tuned too tightly may fail on new images.
• Ignoring threshold tuning: Not adjusting matching threshold can cause poor precision or recall.

Self Check

Your template matching model has 98% accuracy but only 12% recall on real matches. Is it good?

Answer: No, because it misses most real matches (low recall). High accuracy is misleading here since most image areas are non-matches. You should improve recall to find more real matches.

Key Result

Precision and recall are key for template matching; high accuracy alone can be misleading due to many non-match areas.

Practice

1. What is the main purpose of template matching in computer vision?

easy

A. To reduce the size of an image without losing quality

B. To classify images into different categories

C. To find a small image inside a larger image by comparing pixel patterns

D. To generate new images from existing ones

Solution

1. Step 1: Understand template matching concept

   Template matching searches for a smaller image (template) inside a bigger image by comparing pixel patterns.

2. Step 2: Compare with other options

   Other options describe classification, resizing, or generation, which are different tasks.

3. Final Answer:

   To find a small image inside a larger image by comparing pixel patterns -> Option C

4. Quick Check:

   Template matching = find small image inside big image [OK]

Hint: Template matching = locating small image inside big one [OK]

Common Mistakes:

• Confusing template matching with image classification
• Thinking it changes image size
• Assuming it creates new images

2. Which of the following is the correct OpenCV function call to perform template matching?

easy

A. cv2.matchTemplate(image, template, method)

B. cv2.templateMatch(image, template)

C. cv2.findTemplate(image, template, method)

D. cv2.match(image, template)

Solution

1. Step 1: Recall OpenCV template matching syntax

   The correct function is cv2.matchTemplate with parameters (image, template, method).

2. Step 2: Check other options for correctness

   Other options use incorrect function names or missing parameters.

3. Final Answer:

   cv2.matchTemplate(image, template, method) -> Option A

4. Quick Check:

   OpenCV template matching = cv2.matchTemplate [OK]

Hint: Remember exact OpenCV function name: matchTemplate [OK]

Common Mistakes:

• Using wrong function names like templateMatch or findTemplate
• Omitting the method parameter
• Confusing with other OpenCV functions

3. Given the following code snippet, what will be the shape of the result from cv2.matchTemplate(image, template, cv2.TM_CCOEFF_NORMED) if image is 100x100 pixels and template is 20x20 pixels?

medium

A. (100, 100)

B. (81, 81)

C. (120, 120)

D. (80, 80)

Solution

1. Step 1: Understand output size formula

   The output size is (W - w + 1, H - h + 1) where W,H are image dims and w,h are template dims.

2. Step 2: Calculate output shape

   For image 100x100 and template 20x20, output = (100-20+1, 100-20+1) = (81, 81).

3. Final Answer:

   (81, 81) -> Option B

4. Quick Check:

   Output shape = (image - template + 1) [OK]

Hint: Output shape = image size minus template size plus one [OK]

Common Mistakes:

• Using image size directly as output shape
• Adding template size instead of subtracting
• Off-by-one errors in calculation

4. You run template matching but get an error: cv2.error: (-215:Assertion failed) src.type() == templ.type() in function 'matchTemplate'. What is the most likely cause?

medium

A. The template image and source image have different data types or channels

B. The template image is larger than the source image

C. The method parameter is missing in the function call

D. The images are not converted to grayscale

Solution

1. Step 1: Analyze error message

   The error says src.type() == templ.type() failed, meaning image and template types differ.

2. Step 2: Identify cause

   Different data types or number of channels (e.g., one grayscale, one color) cause this error.

3. Final Answer:

   The template image and source image have different data types or channels -> Option A

4. Quick Check:

   Image and template must have same type [OK]

Hint: Check image and template have same type and channels [OK]

Common Mistakes:

• Assuming template size causes this error
• Forgetting to pass method parameter causes this error
• Thinking grayscale conversion is mandatory for all cases

5. You want to detect a rotated version of a template inside an image using template matching. Which approach is best to improve detection?

hard

A. Resize the image to match the template size

B. Use the original template only without rotation

C. Convert both images to grayscale before matching

D. Rotate the template at multiple angles and run template matching for each

Solution

1. Step 1: Understand template matching limitation

   Template matching works best when template matches image exactly in size and orientation.

2. Step 2: Handle rotation

   To detect rotated templates, rotate the template at different angles and match each rotated version.

3. Final Answer:

   Rotate the template at multiple angles and run template matching for each -> Option D

4. Quick Check:

   Rotate template for rotated detection [OK]

Hint: Try multiple rotated templates to detect rotated objects [OK]

Common Mistakes:

• Using only original template ignores rotation
• Resizing image does not fix rotation mismatch
• Grayscale conversion helps but doesn't solve rotation