Template Matching

Template matching is used to locate the occurrence of a smaller image (called a template) within a larger image

Finding a part of an image that matches the given template

result = cv2.matchTemplate(image, templ, method)

• image: The source image (input image) where you want to search for the template. It can be a single-channel (grayscale) or multi-channel (color) image.
• templ:The template image that you want to match against the source image.
• It must be of the same type as the source image and smaller in size.
• method:
• The method used for matching, which can be one of the following constants:
  • cv2.TM_CCOEFF
  • cv2.TM_CCOEFF_NORMED
  • cv2.TM_CCORR
  • cv2.TM_CCORR_NORMED
  • cv2.TM_SQDIFF
  • cv2.TM_SQDIFF_NORMED

The result matrix will be of dimension:

• Width:
  • result_width=image_width−template_width+1
• Height:
  • result_height=image_height−template_height+1

For a template of size (template_width, template_height):

• The maximum x-coordinate for the top-left corner of the template is image_width – template_width.
• The maximum y-coordinate is image_height – template_height.

Original Image: Width = 1025, Height = 1367

Template: Width = 486, Height = 375

The valid top-left corner positions for the template would range:

• X-coordinates: 0 to 1025 – 486 = 539
• Y-coordinates: 0 to 1367 – 375 = 992

This means:

• The template can be placed starting from (0, 0) up to (539, 992) in the original image without any part of it exceeding the original image boundaries.

cv2.TM_SQDIFF

SQDIFF stands for Squared Differences, and it’s a method used in template matching in OpenCV (cv2.TM_SQDIFF)

• Lower values are better: The best match occurs where the squared differences are the smallest, meaning the template and the image region are most similar.

cv2.TM_SQDIFF_NORMED

Normalized Squared Difference

Lower values are better: The best match occurs are the smallest, meaning the template and the image region are most similar.

cv2.TM_CCORR

Cross Correlation – relies on similarity

Higher values are better: A higher score indicates a better match, meaning that the template and the image region are more similar.

cv2.TM_CCORR_NORMED

Normalized Cross Correlation

Higher values are better: A higher score indicates a better match, meaning that the template and the image region are more similar.

Higher values (closer to 1) indicate a better match.

cv2.TM_CCOEFF

Cross Coefficient

• Higher values indicate a better match:
  • Positive values represent good matches, and
  • negative values represent poor matches.

cv2.TM_CCOEFF_NORMED

Normalized Cross Coefficient

• If the normalized score is close to 1, the image region closely matches the template.
• If the score is close to -1, the template is almost a negative (inverted) of the image region.
• A score of 0 means there is no correlation between the template and the image region.

Output returned by minMaxLoc( )

Views: 1