opencv+python (3)

14. Grayscale processing

# 图像灰度化的通道数变化
import numpy as np
img1 = cv2.imread('car.jpg')
print(img1.shape, img1.size, img1.dtype)
img = cv2.imread('car.jpg',cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (300,300))
img1 = cv2.resize(img1, (300,300))
print(img1.shape, img1.size, img1.dtype)
img11 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
print(img11.shape, img11.size, img11.dtype)
print(img11[:, :5])
print(img.shape, img.size, img.dtype)
print(img[:, :5])
cv2.imshow('0', img1)
cv2.imshow('1', img11)
cv2.waitKey(0)
cv2.destroyAllWindows()

15. Canny edge detection

 Canny edge detection:
1) Use a Gaussian filter to smooth the image and filter out noise.

2) Calculate the gradient strength and direction of each pixel in the image.

3) Apply Non-Maximum Suppression suppression to eliminate spurious responses from edge detection.

4) Apply double-threshold (Double-Threshold) detection to determine real and potential edges

5) Edge detection is finally completed by suppressing isolated weak edges

# Canny边缘检测
import numpy as np
img = cv2.imread('car.jpg', cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (300,300))
v1 = cv2.Canny(img, 80, 150)
v2 = cv2.Canny(img, 50, 100)
res = np.hstack((img, v1, v2))
cv2.imshow('0', res)
cv2.waitKey(0)
cv2.destroyAllWindows()

 The upper part is the rendering of gradient calculation, and the lower part is the rendering of canny edge detection

16. Image Pyramid

Pyramid: feature extraction on the image (double the zoom in and zoom out)

1. Gaussian Pyramid

Downsampling

 upsampling

# 高斯金字塔
import numpy as np
img = cv2.imread('car.jpg', cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (300,300))
cv2.imshow('img', img)
# 向上采样
up = cv2.pyrUp(img)
cv2.imshow('up', up)
# 向下采样
down = cv2.pyrDown(img)
cv2.imshow('down', down)
# 先向上再向下
ud = cv2.pyrDown(up)
# 先向下再向上
du = cv2.pyrUp(down)
res = np.hstack((img, ud, du))
cv2.imshow('res', res)
cv2.waitKey(0)
cv2.destroyAllWindows()

 

 

 2. Laplace Pyramid

result = original image - up(down)

 

# 拉普拉斯金字塔
import numpy as np
img = cv2.imread('car.jpg', cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (300,300))
up = cv2.pyrUp(img)
# 先向上再向下
ud = cv2.pyrDown(up)
# 原始图像 - 先向上再向下
result = img - ud
res = np.hstack((img, result))
cv2.imshow('res', res)
cv2.waitKey(0)
cv2.destroyAllWindows()

 Seventeen, contour detection

The difference between the contour and the edge is: the contour is connected together, while the edge can be scattered points, line segments

Contour features, contour approximation

 

 

# 图像轮廓检测
import numpy as np
img = cv2.imread('ellipse.png')
img = cv2.resize(img, (300,300))
# 为了更高的准确率，使用二值化图像
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)   # 灰度化
ret, thresh = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY)   # 二值化
# 找出轮廓
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)   # 新版返回两个值
# binary, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE) # 旧版opencv返回三个值（二值化结果，轮廓信息（list类型），层级）
draw_img = img.copy()   # 复制一个新对象，不然轮廓会画在原图像上面
# 画出轮廓，参数（传入绘制图像，轮廓信息，轮廓索引（-1代表绘制出全部轮廓），颜色模式（BGR），线条厚度）
res = cv2.drawContours(img, contours, -1, (0, 0, 255), 2)
# 取第一个轮廓
cnt = contours[0]
# print(contours)
# 求面积
print(cv2.contourArea(cnt))
# 求周长
print(cv2.arcLength(cnt, True))
# 轮廓近似
epsilon = 0.1*cv2.arcLength(cnt, True)
approx = cv2.approxPolyDP(cnt, epsilon, True)
draw_img1 = img.copy()
res1 = cv2.drawContours(draw_img1, [approx], -1, (0, 0, 255), 2)
cv2.imshow('res1', res1)
cv2.waitKey(0)
cv2.destroyAllWindows()