人脸检测

Haar级联分类器

正常人的脸一定具备眼睛、鼻子、嘴巴等特征，每个特征都做成一个专门的检测分类器，所有分类器串起来，全部检测通过则判定为人脸。

分类器下载地址：

https://github.com/opencv/opencv/tree/master/data/haarcascades

import cv2

#读取待检测的图像
image = cv2.imread("./111.jpg")
cv2.imshow("image",image)

# 转灰度图
gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

# 获取级联分类器对象
faceCascade = cv2.CascadeClassifier('./haarcascades/haarcascade_frontalface_default.xml')

# 检测
faces = faceCascade.detectMultiScale(gray,1.3,3)

print("发现{}个人脸!".format(len(faces)))

#绘制人脸框
for(x,y,w,h) in faces:
    cv2.rectangle(image,(x,y),(x+w,y+w),(0,0,255),2)

#显示
cv2.imshow("result",image)
cv2.waitKey(0)

运行结果：

SSD检测器

SSD检测器是基于深度学习一种检测方式，通过卷积神经网络对图像特征进行检测。

模型下载地址：

https://github.com/opencv/opencv/blob/master/samples/dnn/face_detector/weights.meta4

配置文件下载地址：

https://github.com/opencv/opencv/tree/master/samples/dnn/face_detector

import cv2

#载入模型
model_file = "res10_300x300_ssd_iter_140000_fp16.caffemodel"
config_file = "deploy.prototxt"
net = cv2.dnn.readNetFromCaffe(config_file,model_file)
threshold = 0.9

#读取待检测的图像
img = cv2.imread("./112.jpg")
cv2.imshow('img',img)

frameHeight = img.shape[0]
frameWidth = img.shape[1]

#预处理
blob = cv2.dnn.blobFromImage(img,0.8,(505,505),[104,117,123],False,False)

#检测
net.setInput(blob)
detections = net.forward()

#标记人脸区域
for i in range(detections.shape[2]):
    detection_score = detections[0,0,i,2]
    if detection_score > threshold:
        x1 = int(detections[0,0,i,3] * frameWidth)
        y1 = int(detections[0,0,i,4] * frameHeight)
        x2 = int(detections[0,0,i,5] * frameWidth)
        y2 = int(detections[0,0,i,6] * frameHeight)
        cv2.rectangle(img,(x1,y1),(x2,y2),(0,0,255),2)

cv2.imshow('result',img)
cv2.waitKey(0)

运行结果：

Dlib检测器

dlib依赖于cmake和scikit-image，先把依赖装了再装dlib.

import cv2
import dlib

#读取待检测的图像
img = cv2.imread("./timg.jpg")
cv2.imshow('img',img)

#获取检测器
detector = dlib.get_frontal_face_detector()

#检测
faceRects = detector(img,1)

#标记人脸区域
for faceRect in faceRects:
    x1 = faceRect.left()
    y1 = faceRect.top()
    x2 = faceRect.right()
    y2 = faceRect.bottom()
    cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)

cv2.imshow('result',img)
cv2.waitKey(0)

运行结果：