程式語言：Python
Package：opencv-python
官方網址
OpenCV-Python Tutorials
功能：找出圖片中類似的東西
演算法
- 輸入兩張影像，分別為 image、template
- 不斷滑動 template，得到 image 上各個位置的比較值，比較值代表相似程度
然後將 image 左上角位置，作為 result 比較值的存放位置
- 完成後可得到 result
可用 minMaxLoc() 函式，找出結果圖的最大或最小值，定位出搜尋位置
限制
result = cv2.matchTemplate(image, templ, method[, result])
- 參數
- image
- 被尋找的圖片
- 必須為 8-bit or 32-bit
- templ
- 尋找的物品圖片
- size 不能大於 image，且格式需一致
- method
- result
- 比較的結果，格式為 numpy.ndarray (dtype=float32)
- 可傳入想儲存結果的 array
- 因 image 大小為 \(W \times H\) 且 templ 為 \(w \times h\) ，所以大小為 \((W-w+1) \times (H-h+1)\)
比對方法
\(I\) 表示 image，\(T\) 表示 template，\(R\) 表示 result
- CV_TM_SQDIFF
- 平方差，越小越相似
- \(R(x,y)= \sum _{x',y'} (T(x',y')-I(x+x',y+y'))^2\)
- CV_TM_SQDIFF_NORMED
- 正規化平方差，越小越相似
- 保證當 pixel 亮度都乘上同一係數時，相似度不變
- \(R(x,y)= \frac{\sum_{x',y'} (T(x',y')-I(x+x',y+y'))^2}{\sqrt{\sum_{x',y'}T(x',y')^2 \cdot \sum_{x',y'} I(x+x',y+y')^2}}\)
- CV_TM_CCORR
- 相關係數，越大越相似
- \(R(x,y)= \sum _{x',y'} (T(x',y') \cdot I(x+x',y+y'))\)
- CV_TM_CCORR_NORMED
- 正規化相關係數，越大越相似
- 保證當 pixel 亮度都乘上同一係數時，相似度不變
- \(R(x,y)= \frac{\sum_{x',y'} (T(x',y') \cdot I(x+x',y+y'))}{\sqrt{\sum_{x',y'}T(x',y')^2 \cdot \sum_{x',y'} I(x+x',y+y')^2}}\)
- CV_TM_CCOEFF
- 去掉直流成份的相關係數，越大越相似
- \(R(x,y)= \sum _{x',y'} (T'(x',y') \cdot I'(x+x',y+y'))\)
where
\(\begin{array}{l} T'(x',y')=T(x',y') - 1/(w \cdot h) \cdot \sum
_{x'',y''} T(x'',y'') \\ I'(x+x',y+y')=I(x+x',y+y') - 1/(w \cdot h)
\cdot \sum _{x'',y''} I(x+x'',y+y'') \end{array}\)
- CV_TM_CCOEFF_NORMED
- 正規化 去掉直流成份的相關係數
- 保證當 pixel 亮度都乘上同一係數時，相似度不變
- 計算出的相關係數被限制在了 -1 到 1 之間
- 1 表示完全相同
- -1 表示亮度正好相反
- 0 表示没有線性相關
- \(R(x,y)= \frac{ \sum_{x',y'} (T'(x',y') \cdot I'(x+x',y+y')) }{ \sqrt{\sum_{x',y'}T'(x',y')^2 \cdot \sum_{x',y'} I'(x+x',y+y')^2} }\)
where
\(\begin{array}{l} T'(x',y')=T(x',y') - 1/(w \cdot h) \cdot \sum
_{x'',y''} T(x'',y'') \\ I'(x+x',y+y')=I(x+x',y+y') - 1/(w \cdot h)
\cdot \sum _{x'',y''} I(x+x'',y+y'') \end{array}\)
範例
程式碼
import cv2
import numpy as np
from matplotlib import pyplot as plt
img = cv2.imread('image.png')
img2 = img.copy()
template = cv2.imread('template.jpg')
w = template.shape[1]
h = template.shape[0]
# All the 6 methods for comparison in a list
methods = ['cv2.TM_CCOEFF', 'cv2.TM_CCOEFF_NORMED', 'cv2.TM_CCORR',
'cv2.TM_CCORR_NORMED', 'cv2.TM_SQDIFF', 'cv2.TM_SQDIFF_NORMED']
for meth in methods:
img = img2.copy()
method = eval(meth)
# Apply template Matching
res = cv2.matchTemplate(img,template,method)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
# If the method is TM_SQDIFF or TM_SQDIFF_NORMED, take minimum
if method 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)
cv2.rectangle(img,top_left, bottom_right, 255, 2)
plt.subplot(121),plt.imshow(res,cmap = 'gray')
plt.title('Matching Result'), plt.xticks([]), plt.yticks([])
plt.subplot(122),plt.imshow(img)
plt.title('Detected Point'), plt.xticks([]), plt.yticks([])
plt.suptitle(meth)
plt.show()
結果
程式碼
import cv2
import numpy as np
from matplotlib import pyplot as plt
def getPoints(img, template, threshold, method=cv2.TM_CCOEFF_NORMED):
result = cv2.matchTemplate(img, template, method)
if method in [cv2.TM_SQDIFF, cv2.TM_SQDIFF_NORMED]:
loc = np.where(result <= threshold) #回傳為 y, x
else:
loc = np.where(result >= threshold) #回傳為 y, x
pts = zip(*loc[::-1])
return removeSame(pts, min(template.shape[0], template.shape[1]))
# 去掉太過接近的座標
def removeSame(pts, threshold):
elements = []
for x,y in pts:
for ele in elements:
if ((x-ele[0])**2 + (y-ele[1])**2) < threshold**2:
break
else:
elements.append((x,y))
return elements
def drawRectangle(img, pts, w, h, color=(0,0,255), lineW=2):
for pt in pts:
cv2.rectangle(img, pt, (pt[0] + w, pt[1] + h), color, lineW) #因後面會反向顏色順序，所以這邊顯示紅色要反向
if __name__ == '__main__':
img = cv2.imread('image.png')
img_result = img.copy()
template = cv2.imread('template.png')
w = template.shape[1]
h = template.shape[0]
threshold = 0.4
elements = getPoints(img, template, threshold)
total_n = len(elements)
drawRectangle(img_result, elements, w=template.shape[1], h=template.shape[0])
# 逆時針旋轉 90
template_r90 = cv2.transpose(template)
cv2.flip(template_r90, 0)
pts1 = getPoints(img, template_r90, threshold)
# 順時針旋轉 90
template_r90 = cv2.transpose(template)
cv2.flip(template_r90, 1)
pts2 = getPoints(img, template_r90, threshold)
elements = removeSame(pts1+pts2, w)
total_90 = len(elements)
drawRectangle(img_result, elements, w=template_r90.shape[1], h=template_r90.shape[0], color=(0,255,0))
plt.subplot(121)
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB)) #因 openCV 記錄的顏色順序是反過來的
plt.title('Original'), plt.xticks([]), plt.yticks([])
plt.subplot(122)
plt.imshow(cv2.cvtColor(img_result, cv2.COLOR_BGR2RGB)) #因 openCV 記錄的顏色順序是反過來的
plt.title('Detected Point'), plt.xticks([]), plt.yticks([])
plt.suptitle('Total:{}\nnomral:{}, 90:{}'.format(total_n+total_90, total_n, total_90))
plt.show()
# cv2.imwrite('result.png', img_result)
參考
Template Matching
Template Matching
留言
張貼留言