Template-Matching

Using NCC and Pyramid strategy to deal with the template matching tasks

• Calculate Normalized Correlation Coefficient (NCC)

• Speed up with Pyramid strategy

• Apply median filter to the image for better performance

• Apply power law transform with gamma = 0.4 in special case

• Draw the matching position on the picture

• Calculate the distance between the center position of border box and center pattern ( circle or cross ) in the image.

• Record average time required

  
  

Click to read the code

1.Import library

import os
import cv2
import math
import datetime
import numpy as np
import pandas as pd
import matplotlib.image as img
import matplotlib.pyplot as plt
from google.colab.patches import cv2_imshow

2.Read the images

• Tenplate images are in file : pattern
  • Template_circle
  • Template_cross
  • Template_BorderCircle
  • Template_BorderCross
• Source images are in file : circle or cross
  • Panel{ 1-4 }_circle{ 1-4 }
  • Panel{ 1-4 }_cross{ 1-4 }

3.Template matching

Part 1 : Non-pyramid

Case 1 : Circle

# Read source
circle_ori = cv2.imread('/content/drive/MyDrive/Matching/cross/Panel1_cross2.bmp')
# Apply a "median" blur to the image
blurred = cv2.medianBlur(circle_ori,9)
# Template matching
'''
cir_match(src,temp,temp_border,draw)
src : Source image
temp : Template image
temp_border : Border template image
draw : Image need to be drawn the matching position
'''
circle = cir_match(blurred,temp_cir,temp_border_cir,circle_ori)

Case 2 : Cross

# Read source
cross_ori = cv2.imread('/content/drive/MyDrive/Matching/cross/Panel1_cross2.bmp') 
# Apply a "median" blur to the image
blurred_cro = cv2.medianBlur(cross_ori,9)
# Template matching
'''
cro_match(src,temp,temp_border,draw)
src : Source image
temp : Template image
temp_border : Border template image
draw : Image need to be drawn the matching position
'''
cross = cro_match(blurred_cro,temp_cro,temp_border_cro,cross_ori)

(1) Scan and calculate NCC score through whole source image

(2) Find the highest NCC score and record the corresponding coordinate

(3) Draw the matching position on the image ( Both border box and center pattern )

(4) Calculate the distance between the center position of border box and center pattern ( circle or cross ) in the image

(5) Save all results

Part 2 : Pyramid

Case 1 : Circle

# Read source
circle_ori = cv2.imread('/content/drive/MyDrive/Matching/circle/Panel1_circle1.bmp')

# Apply a "median" blur to the image 
blurred = cv2.medianBlur(circle_ori,9)

# Template matching
'''
cir_match_p(src,temp,temp_border,draw)
src : Source image
temp : Template image
temp_border : Border template image
draw : Image need to be drawn the matching position
'''
circle = cir_match_p(blurred,temp_cir,temp_border_cir,circle_ori)

Case 2 : Cross

# Read source
cross_ori = cv2.imread('/content/drive/MyDrive/Matching/cross/Panel1_cross1.bmp') 
# Apply a "median" blur to the image
blurred_cro = cv2.medianBlur(cross_ori,9)
# Template matching
'''
cro_match_p(src,temp,temp_border,draw)
src : Source image
temp : Template image
temp_border : Border template image
draw : Image need to be drawn the matching position
'''
cross = cro_match_p(blurred_cro,temp_cro,temp_border_cro,cross_ori)

(1) Build pyramid according to the max leval ( default = 10 )

(2) Scan and calculate NCC score through whole source image ( Through all elements in pyramid )

(3) Find the highest NCC score and record the corresponding coordinate

(4) Draw the matching position on the image ( Both border box and center pattern )

(5) Calculate the distance between the center position of border box and center pattern ( circle or cross ) in the image

(6) Save all results

4. Result

• All distance record
• Result : Non-pyramid Circle
• Result : Non-pyramid Cross
• Result : Pyramid Circle
• Result : Pyramid Cross

Part 1 : Non-pyramid

1. Average consume time

• Circle - 0 days 00:24:43.240662375
• Cross - 0 days 00:25:41.053639312

2. Distance

• Circle

Image	Distance
Panel1_circle1	6.324555320336759
Panel1_circle2	5.0990195135927845
Panel1_circle3	3.1622776601683795
Panel1_circle4	13.341664064126334
Panel2_circle1	7.810249675906654
Panel2_circle2	8.06225774829855
Panel2_circle3	2.8284271247461903
Panel2_circle4	11.704699910719626
Panel3_circle1	8.54400374531753
Panel3_circle2	14.142135623730953
Panel3_circle3	17.26267650163207
Panel3_circle4	17.0
Panel4_circle1	4.242640687119285
Panel4_circle2	7.071067811865476
Panel4_circle3	5.656854249492381
Panel4_circle4	13.601470508735444

• Cross

Image	Distance
Panel1_cross1	12.206555615733702
Panel1_cross2	17.08800749063506
Panel1_cross3	16.278820596099706
Panel1_cross4	10.770329614269007
Panel2_cross1	4.47213595499958
Panel2_cross2	10.295630140987
Panel2_cross3	24.166091947189145
Panel2_cross4	11.40175425099138
Panel3_cross1	14.560219778561036
Panel3_cross2	15.811388300841896
Panel3_cross3	3.1622776601683795
Panel3_cross4	15.231546211727817
Panel4_cross1	3.605551275463989
Panel4_cross2	7.0710678118654755
Panel4_cross3	13.416407864998739
Panel4_cross4	5.385164807134504

3. Result image ( show only 1 image from each group )

• Circle

• Cross

Part 2 : Pyramid

1. Average consume time

• Circle - 0 days 00:00:00.749987812
• Cross - 0 days 00:00:00.713385562

2. Distance

• Circle

Image	Distance
Panel1_circle1	6.324555320336759
Panel1_circle2	5.0990195135927845
Panel1_circle3	3.1622776601683795
Panel1_circle4	13.341664064126334
Panel2_circle1	7.810249675906654
Panel2_circle2	8.54400374531753
Panel2_circle3	2.8284271247461903
Panel2_circle4	11.704699910719626
Panel3_circle1	8.54400374531753
Panel3_circle2	14.142135623730951
Panel3_circle3	17.26267650163207
Panel3_circle4	17.0
Panel4_circle1	4.242640687119285
Panel4_circle2	7.810249675906654
Panel4_circle3	10.198039027185569
Panel4_circle4	13.601470508735444

• Cross

Image	Distance
Panel1_cross1	11.40175425099138
Panel1_cross2	16.15549442140351
Panel1_cross3	16.278820596099706
Panel1_cross4	10.770329614269007
Panel2_cross1	4.123105625617661
Panel2_cross2	10.295630140987
Panel2_cross3	23.769728648009426
Panel2_cross4	10.44030650891055
Panel3_cross1	14.560219778561036
Panel3_cross2	15.524174696260024
Panel3_cross3	4.123105625617661
Panel3_cross4	15.231546211727817
Panel4_cross1	3.605551275463989
Panel4_cross2	7.0710678118654755
Panel4_cross3	13.416407864998739
Panel4_cross4	5.385164807134504

3. Result image ( show only 1 image from each group )

• Circle

• Cross

5. Reference

[1] Template matching

https://www.kaggle.com/code/iamarjunchandra/template-matching-by-normalised-cross-correlation

[2] [Python] OpenCV matchTemplate

https://zwindr.blogspot.com/2017/02/python-opencv-matchtemplate.html

[3] opencv+python实现图像匹配

https://blog.csdn.net/zhuisui_woxin/article/details/84400439

[4] Pyramid

https://stackoverflow.com/questions/29605119/fast-template-matching-using-pyramids-in-python