CV-Python核心操作: 图像属性/通道/位运算/色彩空间/缩放/阈值

上一篇：Python OpenCV GUI Features-GUI特征

1. 获取、修改图像像素值

# 获取BGR值
img = cv2.imread('1.jpg')
px = img[100, 100]
print(px) # [157 166 200]

# 获取blue值
blue = img[100,100,0]
print(blue) # 157

------------------以上不推荐（速度慢）------------------
--------------------推荐下面的操作---------------------

# 获取红色值
r = img.item(10,10,2)

# 修改红色值
img.itemset((10,10,2), 100)
print(img.item(10,10,2))  # 100

2. 获取图像基本属性

# 尺寸、通道信息
print(img.shape) # (342, 548, 3)

# 像素数目
print(img.size) # 562248

# 图像类型
print(img.dtype) # uint8

3. ROI: region of interest， 图像中的部分感兴趣区域

roi = img[280:340, 330:390] # 获取roi
img[273:333, 100:160] = roi # 复制到指定区域

4. 分离、合并图像通道

b, g, r = cv2.split(img)
r  =  r // 2   # 红色分量减半
img = cv2.merge(b, g, r)

5. 图像边界（making border / padding）

blue = [255,0,0]
# paras: img, top, bottom, left, right, borderType, value
constant = cv2.copyMakeBorder(img, 10, 10, 10, 10, cv2.BORDER_CONTANT, value=blue)

borderType:
cv2.BORDER_REFLECT: 镜像反射
cv2.BORDER_REFLECT_101 / cv2.BORDER_DEFAULT:  镜像反射
cv2.BORDER_REPLICATE: 复制最后一个元素
cv2.BORDER_WARP: 仿射变换（变形）
cv2.BORDER_CONSTANT: 常量

6. 图像的数学运算（加减乘除）

# 像素值的运算
img = img1 + img2    # <==> img = cv2.add(img1, img2)
img = img1 - img2
img = img1 * img2

7. 图像合成（多张图像相互叠加）

# params: src1, alpha, src2, beta, gamma，alpha和beta为图1图2的比重
dst = cv2.addWeighted(img1, 0.7, img2, 0.3, 0.0)

8. 位运算

img = cv2.imread('1.jpg')  # image
img2 = cv2.imread('2.jpg') # logo
r, c, ch = img2.shape
roi = img[:r, :c]  # 图一左上角，和图二一样大

# 将logo图转换为二值图mask掩膜
grayImg = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY) # 转换为灰度图
ret, mask = cv2.threshold(grayImg, 10, 255, cv2.THRESH_BINARY) # 二值化
mask_inv = cv2.bitwise_not(mask) # 二值图像黑白颠倒

# params: src1, src2, mask
img_bg = cv2.bitwise_and(roi, roi, mask=mask_inv)
img2_fg = cv2.bitwise_and(img2, img2, mask=mask)
dst = cv2.add(img_bg, img2_fg)
img[:r, :c] = dst

9. 色彩空间转换

# bgr -> gray
grayImg = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# bgr -> rgb
rgbImg = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# bgr -> hsv
hsvImg = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)

# bgr -> ycrcb
yccImg = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)


e.g
# demo: track the blue ball in a video
cap = cv2.VideoCapture(0)
while True:
    _, frame = cap.read()
    hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
    lower_blue = np.array([110,50,50])
    upper_blue = np.array([130, 255, 255])
    # 针对hsv空间，更接近人眼感官
    mask = cv2.inRange(hsv, lower_blue, upper_blue)
    res = cv2.bitwise_and(frame, frame, mask=mask)
    cv2.imshow('ball', res)
    if cv2.waitKey(30) == 27:
        break
cv2.destroyAllWindows()

10.  图像变形变换（缩放、翻转、扭曲）

# 缩放 Scaling，放大缩小推荐的插值方式（interpolation）如下：
h, w = img.shape[:2]
img = cv2.resize(img, (h//2,w//2), interpolation=cv2.INTER_CUBIC)     # zoom out
img = cv2.resize(img, (h*2, w*2), interpolation=cv2.INTER_LINEAR)     # zoom in

# 变形 Translation
r, c = img.shape[:2]
M = np.float32([1,0,100], [0,1,50]) # 仿射变换矩阵
dst = cv2.warpAffine(img, M, (c, r))

# 旋转 Rotation
M = cv2.getRotationMatrix2D(((cols-1)/2.0, (rows-1)/2.0), 90, 1)
dst = cv2.warpAffine(img, M, (c, r))

# 仿射变换 Affine Transformation
pts1 = np.float32([[50,50],[200,50],[50,200]])     # src 的四个顶点
pts2 = np.float32([[10,100],[200,50],[100,250]])   # dst 的四个顶点
M = cv2.getRotationMatrix2D(pts1, pts2)
dst = cv2.warpAffine(img, M, (c, r))

# 透视变换 Perspective Transformation
pts1 = np.float32([[56,65],[368,52],[28,387],[389,390]])
pts2 = np.float32([[0,0],[300,0],[0,300],[300,300]])
M = cv2.getPerspectiveTransform(pt1, pt2)
dst = cv2.warpPerpective(img, M, (300, 300))

p.s 仿射变换和透视变换的区别：
仿射变换后还是平行四边形，透视变换后只是四边形了

11. 图像阈值 Threshold

# 最基本的二值化，params: img, min, max, thresholdingType
ret, thresh = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY)

thresholdingType:
cv2.THRESH_BINARY
cv2.THRESH_BINARY_INV
cv2.THRESH_TRUNC
cv2.THRESH_TOZERO
cv2.THRESH_TOZERO_INV
差别对比如下：

# 自适应二值化, params: src, maxVal, adaptiveType, thresholdType, blockSize, C
dst = cv2.adaptiveThreshold(src, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11,2)

adaptiveType：
cv.ADAPTIVE_THRESH_MEAN_C
cv.ADAPTIVE_THRESH_GAUSSIAN_C
差别对比如下：

# opencv自带的二值化算法
cv2.THRESH_BINARY + cv2.OTSU     # 使用大津法自动设置最佳阈值（类内最大方差,基于直方图）
cv2.THRESH_BINARY + cv2.THRESH_TRIANGLE  # 使用三角算法设置最佳阈值

blur = cv2.GaussianBlur(img, (5,5), 0)
ret, thresh = cv2.threshold（img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU）
ret, thresh_guassian = cv2.threshold（blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU）

对比如下：

p.s 以上三张图来自OpenCV官方Documentation。

下一篇：Python OpenCV Image Processing-图像处理