graphics - 使用计算机视觉处理图像验证码

Question

我想在“大图像”内找到“小图像”中对象的坐标。如果您有任何想法，请告诉我。

我尝试过特征匹配，但成功率很低。

---

---

---

Answer 1

对于您的情况，您似乎可以简单地将小图片旋转 N 次，并对每张旋转的小图片应用模板匹配。

还对图像应用中值滤波器以减少噪声

示例:

import cv2 as cv
import numpy as np
from matplotlib import pyplot as plt

img_small = cv.imread('small.png', cv.IMREAD_COLOR)
img_big = cv.imread('big.png', cv.IMREAD_COLOR)

img_small = cv.cvtColor(img_small, cv.COLOR_BGR2RGB)
img_big = cv.cvtColor(img_big, cv.COLOR_BGR2RGB)

img_big = cv.medianBlur(img_big, 3)

rows,cols = img_small.shape[:2]

img_small_rotated = list()

n = 20

for i in range(0,n):
    M = cv.getRotationMatrix2D(((cols-1)/2.0,(rows-1)/2.0),360*(i/n),1)
    img_small_rot = cv.warpAffine(img_small,M,img_small.shape[:2],borderValue = (255,255,255))
    img_small_rot = cv.medianBlur(img_small_rot, 3)
    img_small_rotated.append(img_small_rot)

im_show_all_templates = np.concatenate(img_small_rotated, axis=1)

plt.imshow(im_show_all_templates)
plt.show()

max_loc_global = None
max_val_global = 0

for i in img_small_rotated:
    res = cv.matchTemplate(img_big,i,cv.TM_CCOEFF_NORMED)
    
    min_val, max_val, min_loc, max_loc = cv.minMaxLoc(res)
    if max_val > max_val_global:
        max_val_global = max_val
        max_loc_global = max_loc

print(max_val_global, max_loc_global)

top_left = max_loc_global
bottom_right = (top_left[0] + cols, top_left[1] + rows)

cv.rectangle(img_big,top_left, bottom_right, 0, 2)

plt.imshow(img_big)
plt.show()

结果:

您可以使用类似的技术以其他方式扩展模板匹配空间(缩放、过滤、旋转)，但对于您的应用程序来说，旋转似乎就足够了。