python opencv图像匹配 matchTemplate函数

Bertha 。 2022-04-24 01:00 275阅读 0赞

# 算法思路 #

将目标图像在源图像中滑动，计算两个相同大小图像的距离，最后根据**距离最小**的位置作为匹配结果  
其中计算图像距离的算法有多种：**TM\_SQDIFF**，**TM\_SQDIFF\_NORMED**等[**详细见opencv官网**][opencv]  
***算法很简单，只用用在图像没有发生变形的情况下***

# 代码 #

import cv2
    
    def match_image(source, template):
        # 选择匹配算法
        match_method = cv2.TM_SQDIFF
    
        result = cv2.matchTemplate(source, template, match_method)
        cv2.normalize(result, result, 0, 1, cv2.NORM_MINMAX, -1)
        # 概率图
        cv2.imshow("rate", result)
    
        min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result, None)
        if match_method == cv2.TM_SQDIFF or match_method == cv2.TM_SQDIFF_NORMED:
            match_loc = min_loc
        else:
            match_loc = max_loc
        return match_loc
    
    
    template = cv2.imread("dog.png")
    img = cv2.imread("img.png")
    
    location = match_image(img, template)
    pt1 = (int(location[0]), int(location[1]))
    x = location[0] + img.shape[1] / 2
    y = location[1] + img.shape[0] / 2
    pt2 = (int(x), int(y))
    cv2.rectangle(img, pt1, pt2, (255, 0, 0), 2)
    cv2.imshow("img", img)
    cv2.waitKey(0)

# 匹配结果 #

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NtaWxpZmVf_size_16_color_FFFFFF_t_70]

[opencv]: https://docs.opencv.org/3.3.0/df/dfb/group__imgproc__object.html
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NtaWxpZmVf_size_16_color_FFFFFF_t_70]: /images/20220221/24a21d39e95a4070821d22d989f0def4.png