Content

• Project Description
• Read File
• Adjust template image
• zoom-out
• small-scale calculations
• Calculate similarity
• Define Threshold
• Visualize

---

Project Description

The purpose of this project is to find the template location from the original image, is called Tempalte Matching. In order to speed up the code, the Pyramid Down method will be used.

Read File

There are three types of images read through cv2.imread().

• the target image of color
• the target image of gray
• the template image of gray

Adjust template image

Different target images may not be very similar to the initial template image, possibly due to angular offsets or different image sizes. So rot_image(image, rotation, scale) is to rotate and scale the image.

• Rotation and magnification of each image

  Data	scale	angle
  100-1	X	X
  100-2	X	1°
  100-3	X	X
  100-4	X	X
  Die-1	X	2°
  Die-2	X	-2°

rot_image(img, rot=0, scale=1):

zoom-out

Since the multiple summation of the original template image takes too long and is inefficient, reducing the target image and the template image at the same time through Pyramid Down can greatly shorten the time spent in extracting features.

Data	reduction ratio
100	16
Die	8

small-scale calculations

When we find a high similarity value in the reduced dimensionality image, we need to restore the size of the original image to calculate the similarity in a small range. Since there is an error in the restored image itself, the search range is designed to be set according to the reduction ratio.

get_small_matching_result(img, tmp, points, resize_ratio=8)

Calculate similarity

This project uses Normalized Correlation Coeffiecient(NCC) in Texture Matching to calculate feature similarity.

get_matching_result(img, tmp, output_img):

Define Threshold

Defining appropriate thresholds is one of the important issues for this project.

• The threshold of each image

  Data	threshold
  100-1	0.5
  100-2	0.4
  100-3	0.5
  100-4	0.5
  Die-1	0.79
  Die-2	0.747

Visualize

Finally, use opencv.rectangle() and other visualization tools to output the result, and get the image result after the recognition is successful.

• 100
  • cost time(s)

    Data	My method	OpenCV
    100-1	593.04	0.42
    100-2	595.29	0.41
    100-3	588.19	0.40
    100-4	591.19	0.40

• Die
  • cost time(s)

    Data	My method	OpenCV
    Die-1	204.11	0.39
    Die-2	205.34	0.34