cv2.IMRED_COLOR
:彩色图像cv2.IMREAD_GRAYSCALE
: 灰度图像import cv2
import numpy as np
path = "test.png" # 测试图像
img = cv2.imread(path, cv2.IMREAD_GRAYSCALE) # 默认是彩色图像,可以使用灰度图像
# numpy生成测试图像
# 图像读取函数
def cv_imshow(name, img):
# 图像显示,也可以创建多个窗口
cv2.imshow(name, img)
# 等待时间,毫秒级,0表示任意终止
cv2.waitKey(0)
cv2.destroyAllWindows()
# 保存图像
cv2.imwrite('mytest.png', img) # 保存成功会返回值
# 图像类型
type(img) # numpy.ndarray
# 图像像素点
img.size
# 图像存储类型
img.dtype # dtype('uint8')
# 截取图像--使用索引形式即可
img[0:200, 0:200]
# 颜色通道提取
b, g, r = cv2.split(img)
# 合并
img = cv2.merge((b, g, r))
# 只保留单通道
cur_img = img.copy()
cur_img[:,:,0] = 0 # B 通道置为0
cur_img[:,:.1] = 0 # G 通道置为0
边界填充
# 边界填充
top_size, bottom_size, left_size, right_size = (200,200,200,200)
replicate = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, borderType=cv2.BORDER_REPLICATE) # 复制法,复制边缘像素
reflect = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, borderType=cv2.BORDER_REFLECT) # 反射法,对感兴趣的图像中的像素两边进行复制
reflect101 = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, borderType=cv2.BORDER_REFLECT_101) # 反射法,以最边缘像素为轴 gfedcb|abcdefgh|gfedcba
warp = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, borderType=cv2.BORDER_WRAP) # 外包装法 cdefgh|abcdefgh|abcdefg
constant = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, borderType=cv2.BORDER_CONSTANT, value=0) # 常量法,常数值填充
plt.subplot(231), plt.imshow(img, 'gray'), plt.title('ORIGINAL')
plt.subplot(232), plt.imshow(replicate, 'gray'), plt.title('replicate')
plt.subplot(233), plt.imshow(reflect, 'gray'), plt.title('reflect')
plt.subplot(234), plt.imshow(reflect101, 'gray'), plt.title('reflect101')
plt.subplot(235), plt.imshow(warp, 'gray'), plt.title('warp')
plt.subplot(236), plt.imshow(constant, 'gray'), plt.title('constant')
数值计算
img+10 # 图像每个位置+10
img+img # 相当于(img+img)%256 相同shape对应位置相加
cv2.add(img, img) # 相当于img+img
图像融合
# 两种图像img1(640, 640, 3) img2(320, 320, 3)
cv2.resize(img1, (320, 320)) # resize图像
# cv2.resize(img1, (0, 0), fx=0.5, fy=0.5) # 对图像x,y变成原来的0.5
cv2.addWeighted(img1, 0.4, img2, 0.5, 0) # 0.4*img1 + 0.5*img2 + 0
图像阈值
ret, dst = cv2.threshold(src, thresh, maxval, type)
ret, thresh1 = cv2.threshold(img1, 127, 255, cv2.THRESH_BINARY)
ret, thresh2 = cv2.threshold(img1, 127, 255, cv2.THRESH_BINARY_INV)
ret, thresh3 = cv2.threshold(img1, 127, 255, cv2.THRESH_TRUNC)
ret, thresh4 = cv2.threshold(img1, 127, 255, cv2.THRESH_TOZERO)
ret, thresh5 = cv2.threshold(img1, 127, 255, cv2.THRESH_TOZERO_INV)
titles = ['Original Image', 'BINARY', 'BINARY_INV', 'TRUNC', 'TOZERO', 'TOZERO_INV']
images = [img1, thresh1, thresh2, thresh3, thresh4, thresh5]
for i in range(6):
plt.subplot(2, 3, i+1), plt.imshow(images[i], 'gray')
plt.title(titles[i])
plt.xticks([]), plt.yticks([])
plt.show()
图像滤波
# 均值滤波
# 简单的平均卷积
blur = cv2.blur(img, (3,3))
# 方框滤波
# 基本和均值一样,可以选择归一化
box = cv2.boxFilter(img1, -1, (3,3), normalize=True) # noremalize=True 卷积除以个数,noremal=False 只是卷积求和
# 高斯滤波
# 同一中心点根据距离不同参数的比例不同
gaussian = cv2.GaussianBlur(img1, (5,5), 1)
# 中值滤波
# 相当于用中值代替
median = cv2.medianBlur(img, 5)
res = np.hstack((blur, gaussian, median))
cv_imshow('res', res)
形态学-腐蚀操作
kernel = np.ones((5,5), np.uint8)
erosion = cv2.erode(img, kernel, iterations=1) # iterations 腐蚀次数
形态学-膨胀操作
kernel = np.ones((3, 3), np.uint8)
dige_dilate = cv2.dilate(img, kernel, iterations=1)
开运算与闭运算
# 开:先腐蚀,再膨胀
# 将毛刺去掉
kernel = np.ones((5,5), np.uint8)
opening = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)
# 闭: 先膨胀,再腐蚀
# 扩张毛刺
kernel = np.ones((5,5), np.uint8)
opening = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
梯度运算
# 梯度=膨胀-腐蚀
kernel = np.ones((7,7), np.uint8)
gradient = cv2.morphologyEx(pie, cv2.MORPH_GRADIENT, kernel) # 原图-腐蚀,默认iteration = 1
礼帽与黑帽
# 礼帽
kernel = np.ones((5,5), np.uint8)
tophat = cv2.morphologyEx(img, cv2.MORPH_TOPHAT, kernel)
# 黑帽
tophat = cv2.morphologyEx(img, cv2.MORPH_BLACKHAT, kernel)
图像梯度-Sobel算子
dst = cv2.Sobel(src, ddepth, dx, dy, ksize)
ddepth
: 图像的深度dx和dy分别表示水平和竖直方向
ksize
是Sobel
算子的大小sobelx = cv2.Sobel(img, cv2.CV_64F, 1, 0, ksize=3) # 水平梯度
# 白到黑是正数,黑到白是负数,负数会被截断成0,所以要取绝对值,方法如下:
sobelx = cv2.convertScaleAbs(sobelx)
sobelx = cv2.Sobel(img, cv2.CV_64F, 0, 1, ksize=3) # 垂直梯度
# 白到黑是正数,黑到白是负数,负数会被截断成0,所以要取绝对值,方法如下:
sobelx = cv2.convertScaleAbs(sobelx)
# 分别计算完x, y求和
sobelxy = cv2.addWeighted(sobelx, 0.5, sobely, 0.5, 0)
# 可以直接计算,但是不建议,效果不如分开合起来计算的好
sobelx = cv2.Sobel(img, cv2.CV_64F, 1, 1, ksize=3) # 水平垂直梯度
图像梯度-Scharr算子
scharrx = cv2.Scharr(img, cv2.CV_64F, 1, 0)
scharry = cv2.Scharr(img, cv2.CV_64F, 0, 1)
scharrx = cv2.covertScaleAbs(scharrx)
scharry = cv2.covertScaleAbs(scharry)
scharrxy = cv2.addWeighted(scharrx, 0.5, scharry, 0.5, 0)
图像梯度-laplacian算子
laplacian = cv2..Laplaciap(img, cv2.CV_64F)
laplacian = cv2.converScaleAbs(laplacian)
1. 高斯滤波器
H = [ 0.0924 0.1192 0.0924 0.1192 0.1538 0.1192 0.0924 0.1192 0.0924 ] < − − − 归一化处理 H = \begin{bmatrix} 0.0924&0.1192&0.0924\\ 0.1192&0.1538&0.1192\\ 0.0924&0.1192&0.0924\\ \end{bmatrix}<---归一化处理 H= 0.09240.11920.09240.11920.15380.11920.09240.11920.0924 <−−−归一化处理
e = H ∗ A = [ h 11 h 12 h 13 h 21 h 22 h 23 h 31 h 32 h 33 ] ∗ [ a b c d e f g h i ] = s u m ( [ a × h 11 b × h 12 c × h 13 d × h 21 e × h 22 f × h 23 g × h 31 h × h 32 i × h 33 ] ) e = H*A = \begin{bmatrix} h_{11}&h_{12}&h_{13}\\ h_{21}&h_{22}&h_{23}\\ h_{31}&h_{32}&h_{33}\\ \end{bmatrix}*\begin{bmatrix} a&b&c\\ d&e&f\\ g&h&i\\ \end{bmatrix}=sum(\begin{bmatrix} a\times h_{11}&b\times h_{12}&c\times h_{13}\\ d\times h_{21}&e\times h_{22}&f\times h_{23}\\ g\times h_{31}&h\times h_{32}&i\times h_{33}\\ \end{bmatrix}) e=H∗A= h11h21h31h12h22h32h13h23h33 ∗ adgbehcfi =sum( a×h11d×h21g×h31b×h12e×h22h×h32c×h13f×h23i×h33 )
2. 梯度和方向(Sobel)
G = G x 2 + G y 2 G=\sqrt{G_x^2+G_y^2} G=Gx2+Gy2
θ = a r c t a n ( G y G x ) \theta=arctan(\frac{G_y}{G_x}) θ=arctan(GxGy)
S x = [ − 1 0 1 − 2 0 2 − 1 0 1 ] S y = [ 1 2 1 0 0 0 − 1 − 2 − 1 ] S_x = \begin{bmatrix} -1&0&1\\ -2&0&2\\ -1&0&1\\ \end{bmatrix} S_y = \begin{bmatrix} 1&2&1\\ 0&0&0\\ -1&-2&-1\\ \end{bmatrix} Sx= −1−2−1000121 Sy= 10−120−210−1
G x = S x ∗ A = [ − 1 0 1 − 2 0 2 − 1 0 1 ] ∗ [ a b c d e f h g i ] = s u m ( [ − a 0 c − 2 d 0 2 f − g 0 i ] ) G_x = S_x*A= \begin{bmatrix} -1&0&1\\ -2&0&2\\ -1&0&1\\ \end{bmatrix} * \begin{bmatrix} a&b&c\\ d&e&f\\ h&g&i\\ \end{bmatrix}=sum(\begin{bmatrix} -a&0&c\\ -2d&0&2f\\ -g&0&i\\ \end{bmatrix}) Gx=Sx∗A= −1−2−1000121 ∗ adhbegcfi =sum( −a−2d−g000c2fi )
G y = S y ∗ A = [ 1 2 1 0 0 0 − 1 − 2 − 1 ] ∗ [ a b c d e f h g i ] = s u m ( [ a 2 b c 0 0 0 − g − 2 h − i ] ) G_y = S_y*A= \begin{bmatrix} 1&2&1\\ 0&0&0\\ -1&-2&-1\\ \end{bmatrix} * \begin{bmatrix} a&b&c\\ d&e&f\\ h&g&i\\ \end{bmatrix}=sum(\begin{bmatrix} a&2b&c\\ 0&0&0\\ -g&-2h&-i\\ \end{bmatrix}) Gy=Sy∗A= 10−120−210−1 ∗ adhbegcfi =sum( a0−g2b0−2hc0−i )
v = cv2.Canny(img, 80, 150) # minval = 80, maxval = 150 值越大对于边缘特征提取越细致,信息点过滤越多
高斯金字塔:向上采样方法(放大)
[ 1 4 4 16 ] − − > [ 1 0 4 0 0 0 0 0 4 0 16 0 0 0 0 0 ] \begin{bmatrix} 1&4\\ 4&16\\ \end{bmatrix}-->\begin{bmatrix} 1&0&4&0\\ 0&0&0&0\\ 4&0&16&0\\ 0&0&0&0\\ \end{bmatrix} [14416]−−> 10400000401600000
up = cv2.pyrUp(img)
down = cv2.pyrDown(img)
拉普拉斯金字塔
L i = G i − P y r U P ( P y r D o w n ( G i ) ) L_i=G_i-PyrUP(PyrDown(G_i)) Li=Gi−PyrUP(PyrDown(Gi))
img - cv2.pyrUp(cv2.pyrDown(img))
cv2.findContours(img, mode, method)
mode
:罗阔检索模式
method
:轮廓逼近方法
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY)
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
draw_img = img.copy() # 不拷贝会在img上留下轮廓痕迹
res = cv2.drawContours(draw_img, contours, -1, (0, 0, 255), 2) # -1表示所有轮廓目标,可以0,1,2...对应各种目标
轮廓特征
cnt = contours[0]
# 面积
cv2.contourArea(cnt)
# 周长,True表示闭合的
cv2.arcLength(cnt, True)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY)
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
cnt = contours[0]
# 1. 根据图像形状进行轮廓拟合
epsilon = 0.1*cv2.arcLength(cnt, True) # 周长作为阈值,两点之前使用直线代替的阈值,阈值越小,线段越短,整体性越差
approx = cv2.approxPolyDP(cnt, epsilon, True)
draw_img = img.copy()
res = cv2.drawContours(draw_img, [approx], -1, (0,0,255), 2)
cv_show('res', res)
# 2. 根据边界矩形进行轮廓拟合
x, y, w, h = cv2.boundingRect(cnt)
rec_img = cv2.rectangle(draw_img, (x,y), (x+w, y+h), (0, 255, 0), 2)
cv_show('img', rec_img)
area = cv2.contourArea(cnt)
rect_area = w*h
extent = float(area) / rect_area
print(‘轮廓面积与边界矩形比’, extent)
# 3. 外接圆
(x, y), radius = cv2.minEnclosingCircle(cnt)
center = (int(x), int(y))
radius = int(radius)
rad_img = cv2.circle(draw_img, center, radius, (0, 255, 0), 2)
cv_show('img', rad_img)
模板匹配和卷积原理很像,模板在原图像上从原点开始华东,计算模板与(图像被模板覆盖的地方)的差别程度,这个差别程度的计算方法在opencv里有6种,然后将每次计算的结果放在一个矩阵里,作为结果输出。假设原图形是AxB大小,而模板是axb大小,则输出结果的矩阵是(A-a+1)x(B-b+1)
img = cv2.imread('', 0)
template = cv2.imread('', 0)
h, w = template.shape[:2]
methods = ['cv2.TM_CCOEFF', ‘cv2.TM_CCOEFF_NORMED’, ‘cv2.TM_CCORR’, 'cv2.TM_CCORR_NORMED', 'cv2.TM_SQDIFF', 'cv2.TM_SQDIFF_NORMED']
res = cv2.matchTemplate(img, template, 1, cv2.TM_SQDIFF)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
for meth in methods:
img2 = img.copy()
# 匹配方法的真值
method = eval(meth)
print(method)
res = cv2.matchTemplate(img, template, 1, method)
# 如果是平方差匹配TM_SQDIFF或归一化平方差匹配TM_SQDIFF_NORMED,取最小值
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)
# 画矩形
cv2.rectangle(img2, top_left, bottom_right, 255, 2)
plt.subplot(121), plt.imshow(res, cmap='gray')
plt.xticks([]), plt.yticks([]) # 隐藏坐标轴
plt.subplot(122), plt.imshow(img2, cmap='gray')
plt.xticks([]), plt.yticks([]) # 隐藏坐标轴
plt.suptitle(meth)
plt.show()
匹配多个目标
img = cv2.imread('', 0)
template = cv2.imread('', 0)
h, w = template.shape[:2]
res = cv2.matchTemplate(img, template, 1, cv2.TM_CCOEFF_NORMED)
threshhold = 0.8
# 取匹配程度大于80%的坐标
loc = np.where(res >= threshold)
for pt in zip(*loc[::-1]):
bottom_right = (pt[0] + w, pt[1] + h)
cv2.rectangle(img, pt, bottom_right, (0, 0, 255), 2)
cv_show('img_rgb', img)
cv2.calcHist(images, channels, mask, histSize, ranges)
hist_0 = cv2.calcHist([img], [0], None, [256], [0, 256])
hist_1 = cv2.calcHist([img], [1], None, [256], [0, 256])
hist_2 = cv2.calcHist([img], [2], None, [256], [0, 256])
hist.shape
plt.hist(img_nly.ravel(), 256)
plt.show()
# 或者
color = ('b', 'g', 'r')
for i, col in enumerate(color):
histr = cv2.calcHist([img], [i], None, [256], [0, 256])
plt.plot(histr, color=col)
plt.xlim([0,256])
mask操作
w, h, _ = img.shape
mask = np.zeros(img.shape[:2], np.uint8)
mask[w//2-500:w//2+500, h//2-300:h//2+300] = 255
cv_imshow('mask', mask)
mask_img = cv2.bitwise_and(img_nly, img_nly, mask=mask) # 与操作
cv_imshow("mask_img", mask_img)
hist_full = cv2.calcHist([img], [0], None, [256], [0,256])
hist_mask = cv2.calcHist([img], [0], mask, [256], [0,256])
plt.subplot(221, plt.imshow(img, 'img'))
plt.subplot(222, plt.imshow(mask, 'img'))
plt.subplot(223, plt.imshow(mask_img, 'img'))
plt.subplot(224, plt.plot(hist_full), plt.plot(hist_mask))
plt.xlim([0,256])
plt.show()
# 原图直方图查看
plt.hist(img_nly.ravel(), 256)
plt.show()
# 单通道直方图均衡化
equ = cv2.equalizeHist(img_nly[:,:,0])
plt.hist(equ.ravel(), 256)
plt.show()
# 3通道直方图均衡化
B, G, R = cv2.split(img)
output_B = cv2.equalizeHist(B)
output_G = cv2.equalizeHist(G)
output_R = cv2.equalizeHist(R)
equ = cv2.merge((output_B, output_G, output_R))
plt.hist(equ.ravel(), 256)
plt.show()
# 结果对比
res = np.hstack((img, equ))
cv_imshow('res', res)
自适应直方图均衡化
分块均衡化
# 实例化直方图
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
# 对三通道进行直方图均衡化
B, G, R = cv2.split(img)
res_clahe_B = clahe.apply(B)
res_clahe_G = clahe.apply(G)
res_clahe_R = clahe.apply(R)
res_clahe = cv2.merge((res_clahe_B, res_clahe_G, res_clahe_R))
res = np.hstack((img_nly, equ, res_clahe))
cv_imshow('img', res)
时域和频域的转换
https://zhuanlan.zhihu.com/p/19763358
傅里叶变换作用
滤波
低通滤波器:只保留低频,会使得图像模糊
高通滤波器:只保留高频,会使得图像细节增强
opencv中主要使用cv2.dft()和cv2.idft(),输入图像需要先转换成np.float32格式。
得到的结果中频率为0的部分会在左上角,通常要转换到中心位置,可以通过shift变换来实现。
cv2.dft()返回的结果是双通道的(实部,虚部),通常还需要转换成图像格式才能展示(0,255)。
img_float32 = np.float32(img[:,:,0])
dft = cv2.dft(img_float32, flags = cv2.DFT_COMPLEX_OUTPUT) # 傅里叶变换
dft_shift = np.fft.fftshift(dft) # 移动
magnitude_spectrum = 20*np.log(cv2.magnitude(dft_shift[:,:,0], dft_shift[:,:,1])) # magnitude实部和虚部调整
plt.subplot(121), plt.imshow(img_nly, cmap='gray')
plt.title('Input Image'), plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(magnitude_spectrum, cmap='gray')
plt.title('Magnitude Spectrum'), plt.xticks([]), plt.yticks([])
plt.show()
img_float32 = np.float32(img[:,:,0])
dft = cv2.dft(img_float32, flags = cv2.DFT_COMPLEX_OUTPUT)
dft_shift = np.fft.fftshift(dft)
rows, cols = img_nly.shape[:2]
crow, ccol = int(rows/2), int(cols/2)
# 低通滤波
mask = np.zeros((rows, cols, 2), np.uint8)
mask[crow-30:crow+30, ccol-30: ccol+30] = 1
# IDFT
fshift = dft_shift*mask
f_ishift = np.fft.ifftshift(fshift)
img_back = cv2.idft(f_ishift)
img_back = cv2.magnitude(img_back[:,:,0], img_back[:,:,1])
plt.subplot(121), plt.imshow(img_nly, cmap='gray')
plt.title('Input Image'), plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(img_back, cmap='gray')
plt.title('Magnitude Spectrum'), plt.xticks([]), plt.yticks([])
plt.show()
cv2.VideoCapture可以捕获摄像头,用数字来控制不同的设备,例如0,1
如果是视频文件,直接指定好路径即可
cv2.COLOR_BGR2RGB
将BGR格式转换成RGB格式
cv2.COLOR_BGR2GRAY
将BGR格式转换成灰度图片
cv2.COLOR_BGR2BGRA
将BGR格式转换成BGR格式,cv2显示正常
cv2.COLOR_BGR2HSV
将BGR格式转换成HSV格式
import cv2
import numpy as np
vc = cv2.VideoCapture('test.mp4') # 打开视频
# vc = cv2.VideoCapture(0) # 打开摄像头0
# 判断是否正确读取视频
if vc.isOpened():
open, fram = vc.read()
else:
open = False
# 播放视频
while open:
ret, frame = vc.read()
if frame is None:
break
if ret:
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
cv2.imshow('result', gray)
if cv2.waitKey(10) & 0xFF == 27: # 27退出键
break
vc.release()
cv2.destroyAllWindows()