模板匹配

1import cv2
2import numpy as np
3
4ring_outer_diameter_in_pixels = 26
5ring_inner_diameter_in_pixels = 18
6
7original_img = cv2.imread('./images/electronics/beaglebone.jpg')
8print(f"original_img.shape = {original_img.shape}")
9
10# 创建模板
11template_shapeHWC = [round(1.5 * ring_outer_diameter_in_pixels),
12                        round(1.5 * ring_outer_diameter_in_pixels), 3]
13if template_shapeHWC[0] % 2 == 0:
14    template_shapeHWC[0] += 1
15if template_shapeHWC[1] % 2 == 0:
16    template_shapeHWC[1] += 1
17print(f"template_shapeHWC = {template_shapeHWC}")
18template_img = np.zeros(template_shapeHWC, dtype=np.uint8)
19
20# 绘制环形：先画一个黄色圆盘，再在其上覆盖一个较小的黑色圆盘
21cv2.circle(template_img, center=(template_shapeHWC[1]//2, template_shapeHWC[0]//2), radius=ring_outer_diameter_in_pixels//2,
22            color=(128, 210, 255), thickness=-1)
23cv2.circle(template_img, center=(template_shapeHWC[1] // 2, template_shapeHWC[0] // 2), radius=ring_inner_diameter_in_pixels//2,
24            color=0, thickness=-1)
25            
26cv2.imwrite('./output/0_original.png', original_img)
27cv2.imwrite('./output/1_template.png', template_img)

1# ... 接上文
2
3# 执行模板匹配
4match_img = cv2.matchTemplate(original_img, template_img, method=cv2.TM_CCOEFF_NORMED)
5print(f"match_img.dtype = {match_img.dtype}")
6print(f"match_img.shape = {match_img.shape}")
7
8# 输出匹配值的范围
9minimum_match_value = np.min(match_img)
10maximum_match_value = np.max(match_img)
11print(f"minimum_match_value = {minimum_match_value}; maximum_match_value = {maximum_match_value}")
12
13# 将匹配结果转换为8位图像
14match_8b_img = (127.0 + 128.0 * match_img).astype(np.uint8)
15
16# 添加标注并保存
17annotated_match_8b_img = cv2.cvtColor(match_8b_img, cv2.COLOR_GRAY2BGR)
18cv2.putText(annotated_match_8b_img, 'match', (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 1.0, (255, 0, 0), 2)
19cv2.imwrite('./output/1_match.png', annotated_match_8b_img)

dimensions_{dst} = (height_{source} - height_{template} + 1, width_{source} - width_{template} + 1)

1# ... 接上文
2
3# 对匹配结果进行零填充，使其尺寸与原始图像一致
4zero_padded_match_img = np.zeros((original_img.shape[0], original_img.shape[1]), dtype=float)
5# 将匹配图像复制到零填充图像的中心区域
6zero_padded_match_img[template_shapeHWC[0]//2: template_shapeHWC[0]//2 + match_img.shape[0],
7    template_shapeHWC[1]//2: template_shapeHWC[1]//2 + match_img.shape[1]] = match_img
8
9# 对匹配结果进行阈值处理
10match_threshold = 0.43
11retval, thresholded_match_img = cv2.threshold(zero_padded_match_img, match_threshold, 255, cv2.THRESH_BINARY)
12
13# 在原始图像上做标注
14annotated_img = copy.deepcopy(original_img)
15for y in range(thresholded_match_img.shape[0]):
16    for x in range(thresholded_match_img.shape[1]):
17        if thresholded_match_img[y, x] > 0:
18            annotated_img[y, x] = (0, 0, 255)
19
20# 标注并保存阈值化后的匹配图
21annotated_thresholded_match_img = cv2.cvtColor(thresholded_match_img.astype(np.uint8), cv2.COLOR_GRAY2BGR)
22cv2.putText(annotated_thresholded_match_img, 'thresholded match', (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 1.0, (255, 0, 0), 2)
23cv2.imwrite('./output/2_thresholded_match.png', annotated_thresholded_match_img)
24
25# 标注叠加在原图上的掩膜区域
26cv2.putText(annotated_img, 'active mask (red) over the original image', (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.8, (255, 0, 0), 2)
27cv2.imwrite('./output/3_annotated.png', annotated_img)