如何计算边缘的平均偏差和标准偏差（OpenCV 中的 Canny 边缘检测）

Question

我有一些图像，您可以在其中看到打印的半色调，我想计算这些点的平均值和标准差。我正在使用 OpenCV 和 Canny 边缘检测来隔离这样的点

import cv2

img = cv2.imread('img/color_dots.png')
img_blur = cv2.GaussianBlur(img, (3,3), 0)

edges = cv2.Canny(image=img_blur, threshold1=50, threshold2=100)

这是原图：

这里是精巧的边缘图像：

我认为并不是所有的点都能被识别出来，尤其是黄色的... 而且我不确定我的方法是否适用于边缘检测或更好地找到轮廓？无论如何，我如何计算这些点的平均大小？

Answer 1

这是 Python/OpenCV 中的一种方法。

阈值背景以隔离点。然后，您可以使用遮罩直接在 OpenCV 中计算颜色的平均值。然后可以从方差的平方根得到标准差，它是图像平方的均值减去图像均值的平方。参见 https://en.wikipedia.org/wiki/Standard_deviation

（或者，使用 Numpy 直接从 np.mean 和 np.std 计算均值和标准差）

输入：

import cv2
import numpy as np

img = cv2.imread("color_dots.png")

# threshold on background color
lower = (215,215,215)
upper = (255,255,255)
thresh = cv2.inRange(img, lower, upper)

# invert so dots are white
thresh = 255 - thresh
#thresh = cv2.cvtColor(thresh, cv2.COLOR_GRAY2BGR)

# get mean colors of dots using mask
mean_values = cv2.mean(img, mask=thresh)[0:3]
mean_values = list(mean_values)
print("mean:", mean_values)

# compute square of image (as floats) and compute mean of squared image
imgf = img.astype(np.float64)
imgf2 = imgf * imgf
mean2_values = cv2.mean(imgf2, mask=thresh)[0:3]

# convert mean of image and mean of image squared tuples to arrays
mean_values_arr = np.array([mean_values])
mean2_values_arr = np.array([mean2_values])

# compute the variance from the mean of image and mean of image squared arrays
variance_values_arr = mean2_values_arr - (mean_values_arr)*(mean_values_arr)

# compute sqrt to form std
std_values_arr = np.sqrt(variance_values_arr)

# convert array to simple list
std_values = list(std_values_arr[0])
print("std:", std_values)

# save result
cv2.imwrite("color_dots_threshold.png",thresh)

cv2.imshow("thresh", thresh)
cv2.waitKey(0)
cv2.destroyAllWindows()

阈值图像：

均值和标准差：

mean: [226.75895493644884, 209.73003594813198, 212.56531647194763]
std: [21.657486571574186, 19.53005158116894, 30.588081007741454]

加法

这是使用 Numpy 的另一种更直接的方法。

import cv2
import numpy as np

img = cv2.imread("color_dots.png")

# threshold on background color
lower = (215,215,215)
upper = (255,255,255)
thresh = cv2.inRange(img, lower, upper)

# invert so dots are white
thresh = 255 - thresh

# separate channels 
b,g,r = cv2.split(img)

# compute mean of each channel
bmean = np.mean(b[np.where(thresh==255)])
gmean = np.mean(g[np.where(thresh==255)])
rmean = np.mean(r[np.where(thresh==255)])

# compute std of each channel
bstd = np.std(b[np.where(thresh==255)])
gstd = np.std(g[np.where(thresh==255)])
rstd = np.std(r[np.where(thresh==255)])

# print results
print("mean:", bmean,gmean,rmean)
print("std:", bstd,gstd,rstd)

# save result
cv2.imwrite("color_dots_threshold.png",thresh)

cv2.imshow("thresh", thresh)
cv2.waitKey(0)
cv2.destroyAllWindows()

均值和标准差：

mean: 226.75895493644884 209.73003594813198 212.56531647194763
std: 21.657486571574225 19.53005158116893 30.58808100774145

加法 2

这可能是使用 OpenCV 的最简单方法。

import cv2
import numpy as np

img = cv2.imread("color_dots.png")

# threshold on background color
lower = (215,215,215)
upper = (255,255,255)
thresh = cv2.inRange(img, lower, upper)

# invert so dots are white
thresh = 255 - thresh

# compute mean and standard deviation
mean, std = cv2.meanStd(img, mask=thresh)[0:3]

# print results
print("mean:", mean)
print("std:", std)

# save result
cv2.imwrite("color_dots_threshold.png",thresh)

cv2.imshow("thresh", thresh)
cv2.waitKey(0)
cv2.destroyAllWindows()

均值和标准差：

mean: 226.75895493644884 209.73003594813198 212.56531647194763
std: 21.657486571574225 19.53005158116893 30.58808100774145

如何计算边缘的平均偏差和标准偏差（OpenCV 中的 Canny 边缘检测）

How to compute Mean and Standard Deviation of Edges (Canny Edge Detection in OpenCV)

python

opencv

computer-vision