[opencv learning] [template matching]

Learn template matching today
Template image

Target image

import cv2
import matplotlib.pyplot as plt


#  The principle of template matching ：
#  A template image slides on the target image from left to right and from top to bottom , The matching degree is calculated for each sliding position （ Mean square deviation or correlation coefficient ）
#  After traversing the target image , Find the best matching unknown region .

#  Show the image , Encapsulate as a function 
def cv_show_image(name, img):
    cv2.imshow(name, img)
    cv2.waitKey(0)  #  Waiting time , In milliseconds ,0 Represents any key termination 
    cv2.destroyAllWindows()


#  Read the template 
template = cv2.imread('images/template.jpg')  #  Turn to grayscale 
print('template Image dimension is :', template.shape)
h, w, c = template.shape
print(h)
print(w)
cv_show_image('template', template)

#  Read the target image 
dst_img = cv2.imread('images/saoge2.jpg')  #  Turn to grayscale 
print(' The target image dimension is :', dst_img.shape)
cv_show_image('dst_img', dst_img)

#  The measurement method of matching degree 
methords = ['cv2.TM_SQDIFF',  #  Calculate the mean square deviation of each pixel , The smaller the value, the more relevant 
            'cv2.TM_CCORR',  #  Calculate correlation , The larger the value, the more relevant 
            'cv2.TM_CCOEFF',  #  Calculate the correlation coefficient , The larger the value, the more relevant 
            'cv2.TM_SQDIFF_NORMED',  #  Calculate the mean square deviation of each pixel , The smaller the value, the more relevant , With normalization 
            'cv2.TM_CCORR_NORMED',  #  Calculate correlation , The larger the value, the more relevant , With normalization 
            'cv2.TM_CCOEFF_NORMED'  #  Calculate the correlation coefficient , The larger the value, the more relevant , With normalization 
            ]

colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255), (255, 255, 0), (0, 255, 255), (255, 0, 255)]
thickness = [12, 10, 8, 6, 4, 2]

img_result = dst_img.copy()
for i in range(len(methords)):
    img_tmp = dst_img.copy()  #  Show the current frame temporarily 
    methord = eval(methords[i])  #  Get the specific template matching method 
    print('current methord is: {}'.format(methords[i]))

    #  Actual template matching 
    res = cv2.matchTemplate(dst_img, template, methord)  #  Incoming target image 、 Template image 、 Match metrics 
    print(res.shape)  #  Get is  (dst_img.H - template.H + 1, dst_img.W - template.W + 1) Dimension data of 
    print(type(res))  # <class 'numpy.ndarray'>
    print(res.dtype)  #  The type of each element is  float32

    #  Analysis results , Obtain the maximum and minimum values of the measurements in the matching process respectively , And the location of its distribution pixels , according to methord Different , Take the appropriate value 
    min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
    if methord in [cv2.TM_SQDIFF, cv2.TM_SQDIFF_NORMED]:
        top_left = min_loc
    else:
        top_left = max_loc
    bottom_right = (top_left[0] + w, top_left[1] + h)
    print(' The location information that the template matches is ', top_left, bottom_right)

    #  Draw a rectangle on the image to indicate the matching position 
    cv2.rectangle(img_result, top_left, bottom_right, colors[i], thickness[i])
    cv2.rectangle(img_tmp, top_left, bottom_right, colors[i], thickness[i])

    plt.subplot(1, 2, 1), plt.imshow(res, 'gray')  #  Draw each pixel （ Representing one (h, w) Region ） The matching degree of 
    plt.xticks([]), plt.yticks([])  #  Hide axis 
    plt.subplot(1, 2, 2), plt.imshow(img_tmp, 'gray')  #  Draw the matching position on the original image 
    plt.xticks([]), plt.yticks([])  #  Hide axis 
    plt.title(methords[i])
    plt.show()

#  Finally, check all the frames together 
cv_show_image('img_result', img_result)

The effect is as follows ：
‘cv2.TM_SQDIFF’, # Calculate the mean square deviation of each pixel , The smaller the value, the more relevant

‘cv2.TM_CCORR’, # Calculate correlation , The larger the value, the more relevant

‘cv2.TM_CCOEFF’, # Calculate the correlation coefficient , The larger the value, the more relevant

‘cv2.TM_SQDIFF_NORMED’, # Calculate the mean square deviation of each pixel , The smaller the value, the more relevant , With normalization

‘cv2.TM_CCORR_NORMED’, # Calculate correlation , The larger the value, the more relevant , With normalization

‘cv2.TM_CCOEFF_NORMED’ # Calculate the correlation coefficient , The larger the value, the more relevant , With normalization

The final result is