OpenCv-色彩域

常见色彩空间

• RGB
• HSV
• HIS
• YCrCb
• YUV

色彩空间转换

• 最常见的两个

• HSV与RGB
• YUV与RGB

代码练习3（颜色空间转换，通道分割，色彩追踪）：

import cv2 as cv
import numpy as np

image_path = "datasources/images/1 (1).jpg"
vido_path = "E:\\视频剪辑素材\\yizhan.mp4"
def replace_color(image):
src = cv.imread(image)

#cv.cvtColor颜色空间转换
HSV = cv.cvtColor(src, cv.COLOR_BGR2HSV)
cv.imshow("HSV", HSV)
HLS = cv.cvtColor(src, cv.COLOR_BGR2HLS)
cv.imshow("HIS", HLS)
YCrCb = cv.cvtColor(src, cv.COLOR_BGR2YCrCb)
cv.imshow("YCrCb", YCrCb)
YUV = cv.cvtColor(src, cv.COLOR_BGR2YUV)
cv.imshow("YUV", YUV)

cv.waitKey(0)
cv.destroyAllWindows()

def colcor_track(vido):
caputer = cv.VideoCapture(vido)
while(True):
ret, frame = caputer.read()
if ret==False:
break
#原图RGB转换为HSV
hsv = cv.cvtColor(frame, cv.COLOR_BGR2HSV)
hmain = np.array([35, 43, 46])
hmax = np.array([77, 255, 255])
msk = cv.inRange(hsv, hmain, hmax)
cv.imshow("yizhan_video", frame)
cv.imshow("mask_vedo", msk)

c = cv.waitKey(40)
if c==27:
break

def split_iamge(image):

src = cv.imread(image, 1)
# 分隔bgr通道
b, g, r = cv.split(src)
cv.imshow("b", b)
cv.imshow("g", g)
cv.imshow("r", r)

src[:, :, 0] = 255
cv.imshow("change_channels2", src)
#合并通道
new_image = cv.merge([b, g, r])
cv.imshow("new_imge", new_image)
cv.waitKey(0)
cv.destroyAllWindows()

if __name__ =="__main__":
# replace_color(image_path)
# colcor_track(vido_path)
split_iamge(image_path)

人脸检测

# -*- coding=GBK -*-
import cv2 as cv
import numpy as np
import os
image_path = "../personimage/AFLW/0004-image64929.jpg"
xml_path = "../opencvcore/opencv/data/haarcascades/haarcascade_frontalface_alt_tree.xml"
video_path = "E:\\视频剪辑素材\\lzlh.ts"

def person_face_detect(image, xml):

src = cv.imread(image)
#和后面的imshow名字对应
cv.namedWindow("input image", cv.WINDOW_AUTOSIZE)
cv.namedWindow("result", cv.WINDOW_AUTOSIZE)
cv.imshow('input image', src)
#转换成灰度图像
gray = cv.cvtColor(src, cv.COLOR_BGR2GRAY)
#级联检测器
face_detector = cv.CascadeClassifier(xml)
#在多个尺度空间进行人脸检测，尺度变化，识别度越打越高
faces = face_detector.detectMultiScale(gray, 1.02, 5)
#绘制人脸框
for x, y, w, h in faces:
cv.rectangle(src, (x, y), (x+w, y+h), (0, 0, 255), 2)
cv.imshow("result", src)

cv.waitKey(0)
cv.destroyAllWindows()

def face_interfac(src, xml):

#和后面的imshow名字对应
# cv.namedWindow("input video", cv.WINDOW_AUTOSIZE)
cv.namedWindow("result", cv.WINDOW_AUTOSIZE)
# cv.imshow('input video', src)
#转换成灰度图像
gray = cv.cvtColor(src, cv.COLOR_BGR2GRAY)
#级联检测器
face_detector = cv.CascadeClassifier(xml)
#在多个尺度空间进行人脸检测，尺度变化，识别度越打越高
faces = face_detector.detectMultiScale(gray, 1.02, 5)
#绘制人脸框
for x, y, w, h in faces:
cv.rectangle(src, (x, y), (x+w, y+h), (0, 0, 255), 2)
cv.imshow("result", src)

#视频中的人脸识别
def person_face_video():
caputer = cv.VideoCapture(video_path)
while(True):
ret, frame = caputer.read()
if ret==False:
break

face_interfac(frame, xml_path)
if cv.waitKey(1)==0:
break

if __name__ == "__main__":
person_face_detect(image_path, xml_path)
# person_face_video()

验证码识别（tesseract-OCR+pytesseract ）

import cv2 as cv
from PIL import Image
import pytesseract

image_path = "datasources/yzm/a510513a-7bc6-4217-bfb3-ad3c197c0e3d.png"

def recognize_text():
src = cv.imread(image_path, 1)
cv.imshow("原来", src)
gray = cv.cvtColor(src, cv.COLOR_BGR2GRAY)
#获得二值图
ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY_INV | cv.THRESH_OTSU)

#以下的三步binary良好可省略
#获得指定形状的kernel,矩形MORPH_RECT;交叉形：MORPH_CROSS;椭圆形：MORPH_ELLIPSE;
kernel = cv.getStructuringElement(cv.MORPH_RECT, (1, 2))
#先腐蚀进行开运算，去除噪声
binl = cv.morphologyEx(binary, cv.MORPH_OPEN, kernel)
#再次获取kernel的形状的
kernel = cv.getStructuringElement(cv.MORPH_RECT, (2, 1))

open_out = cv.morphologyEx(binl, cv.MORPH_OPEN, kernel)
#二进制像素进行与
cv.bitwise_not(open_out, open_out)
cv.imshow("转换", open_out)
#array转换成为image
textImage = Image.fromarray(open_out)
#OCR识别image中的文字，需要由tesseract包装ocr才可以识别
text = pytesseract.image_to_string(textImage)
print("This OK:%s"%text)

cv.waitKey(0)
cv.destroyAllWindows()

if __name__ == "__main__":
recognize_text()

• 总结(全):

基本完成，边缘提取分三个步骤（1.图像降噪2.计算图像梯度3.非极大值抑制
4.阈值筛选），膨胀和腐蚀可以降低image的噪声，这里的人脸检测其实就是将人脸图像二值化，调用haarcascades的级联检测器实现的，验证码也是通过二值化然后使用Google的pytesseract +ocr实现的。
重点： Canny边缘提取、人脸检测，验证码识别
难点： 膨胀与腐蚀