opencv3/C++/Python 检测分类器使用（HAAR/LBP）:人脸与眼睛的检测、猫脸检测

利用OpenCV自带的xml文件，实时检测摄像头中人脸与眼睛。

opencv3/C++

detectMultiScale()参数说明：

//在输入图像中检测不同大小的对象。检测到的对象作为矩形列表返回。
void detectMultiScale( 
InputArray image,//CV_8U类型图像
CV_OUT std::vector& objects,//包含检测到的对象的矩形
double scaleFactor = 1.1,//指定在每个图像比例下图像大小减少的参数。
int minNeighbors = 3, //指定每个候选矩形必须保留多少个邻居。
int flags = 0,//与旧函数cvHaarDetectObjects中的相同
Size minSize = Size(),//最小可能的对象大小
Size maxSize = Size() //最大可能的对象大小
);

人脸与眼睛的检测示例：

注释掉的部分为绘制眼睛移动的轨迹。

#include
using namespace cv;

//人脸与眼睛检测与定位
int main()  
{  
    CascadeClassifier faceCascader, eyeCascader;
    String filename1 = "D:/opencv3.1.0/opencv/tools/opencv_contrib/install/etc/haarcascades/haarcascade_frontalface_alt.xml";
    String filename2 = "D:/opencv3.1.0/opencv/tools/opencv_contrib/install/etc/haarcascades/haarcascade_eye.xml";   
    //std::vector pt; //绘制轨迹

    if (!faceCascader.load(filename1))
    {
        printf("can not load the face feature data \n");
        return -1;
    }
    if (!eyeCascader.load(filename2))
    {
        printf("can not load the eye feature data \n");
        return -1;
    }
    namedWindow("input", WINDOW_AUTOSIZE);
    VideoCapture capture;
    capture.open(0);
    Mat frame, gray, src;
    std::vector faces, eyes;
    while (capture.read(frame))
    {
        flip(frame, frame, 1);
        src = frame;
        cvtColor(frame, gray, COLOR_BGR2GRAY);
        equalizeHist(gray, gray);
        faceCascader.detectMultiScale(gray, faces,1.2, 3, 0, Size(30, 30));
        for (int i = 0; i < faces.size(); i++)
        {
            Rect roi;
            roi.x = faces[static_cast<int>(i)].x;
            roi.y = faces[static_cast<int>(i)].y;
            roi.width = faces[static_cast<int>(i)].width;
            roi.height = faces[static_cast<int>(i)].height /2.0;
            Mat faceROI = frame(roi);
            eyeCascader.detectMultiScale(faceROI, eyes,1.2, 3, 0, Size(20, 20));
            for (int j = 0; j < eyes.size(); j++)
            {
                Rect rect;
                rect.x = faces[static_cast<int>(i)].x + eyes[j].x + eyes[j].width/2.0;
                rect.y = faces[static_cast<int>(i)].y + eyes[j].y + eyes[j].height/2.0;
                rect.width = eyes[j].width;
                rect.height = eyes[j].height;
                circle(frame, Point(rect.x, rect.y), rect.width/2.0, Scalar(0,0,255), 2, 8);
                //pt.push_back(Point(rect.x,rect.y));  
            }
            rectangle(frame, faces[static_cast<int>(i)], Scalar(0,255,0), 2, 8, 0);
        }
        char q = waitKey(10);
        //Esc退出
        if (q == 27)
        {
            break;
        }
        /*
        //绘制眼睛轨迹
              for(int i=0;i
        imshow("input", frame);
    }
    waitKey(0);
    capture.release();
    return 0;  
}  

猫脸检测

#include
using namespace cv;

int main()  
{  
    CascadeClassifier haarfaceCascader;
    String catfile = "D:/opencv3.1.0/opencv/tools/opencv_contrib/install/etc/haarcascades/haarcascade_frontalcatface_extended.xml";

    Mat src = imread("E:/image/image/cat.jpg");
    Mat gray;
    std::vector faces;
    cvtColor(src, gray, COLOR_BGR2GRAY);
    equalizeHist(gray, gray);

    if (!haarfaceCascader.load(catfile))
    {
        printf("can not load the haar file \n");
        return -1;
    }
    haarfaceCascader.detectMultiScale(gray, faces, 1.1, 3, 0, Size(30, 30), Size(300, 300));


    for (int i = 0; i < faces.size(); i++)
    {
        rectangle(src, faces[i], Scalar(0,255,0), 2, 8, 0);
    }

    namedWindow("input", WINDOW_NORMAL);
    imshow("input",src);

    waitKey(0);
    return 0;
}

HAAR与LBP区别

HAAR与LBP区别：
① HAAR特征是浮点数计算，LBP特征是整数计算；
② LBP训练需要的样本数量比HAAR大；
③ LBP的速度一般比HAAR快；
④ 同样的样本HAAR训练出来的检测结果要比LBP准确；
⑤ 扩大LBP的样本数据可达到HAAR的训练效果。

对比二者用时：

#include
using namespace cv;

int main()  
{  
    CascadeClassifier haarfaceCascader, lbpfaceCascader;
    String haarfile = "D:/opencv3.1.0/opencv/tools/opencv_contrib/install/etc/haarcascades/haarcascade_frontalface_alt.xml";
    String lbpfile = "D:/opencv3.1.0/opencv/tools/opencv_contrib/install/etc/lbpcascades/lbpcascade_frontalface.xml";

    Mat src = imread("E:/image/image/sophie.jpg");
    Mat gray;
    std::vector faces1, faces2;
    cvtColor(src, gray, COLOR_BGR2GRAY);
    equalizeHist(gray, gray);

    int st1 = getTickCount();
    if (!haarfaceCascader.load(haarfile))
    {
        printf("can not load the haar file \n");
        return -1;
    }
    haarfaceCascader.detectMultiScale(gray, faces1, 1.1, 3, 0, Size(30, 30));
    int et1 = (getTickCount() - st1);
    printf("Time to load the haar file is : %d \n", et1);

    int st2 = getTickCount();
    if (!lbpfaceCascader.load(lbpfile))
    {
        printf("can not load the lbp file \n");
        return -1;
    }
    lbpfaceCascader.detectMultiScale(gray, faces2, 1.1, 3, 0, Size(30, 30));
    int et2 = (getTickCount() - st2);
    printf("Time to load the lbp file is : %d \n", et2);
    for (int i = 0; i < faces1.size(); i++)
    {
        rectangle(src, faces1[i], Scalar(0,255,0), 2, 8, 0);
    }
    for (int i = 0; i < faces2.size(); i++)
    {
        rectangle(src, faces2[i], Scalar(0,0,255), 4, 8, 0);
    }
    namedWindow("input", WINDOW_AUTOSIZE);
    imshow("input",src);

    waitKey(0);
    return 0;
}

opencv3/Python

人脸以及眼睛识别:

#!/usr/bin/python
# coding:utf8

import cv2

face_xml = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
face_xml.load('haarcascade_frontalface_default.xml')
eye_xml = cv2.CascadeClassifier('haarcascade_eye.xml')
eye_xml.load('haarcascade_eye.xml')

if __name__ == '__main__':
    cam = cv2.VideoCapture(0)
    # cam = cv2.VideoCapture('broke_girls.mp4')
    # 获得码率及尺寸
    fps = cam.get(cv2.CAP_PROP_FPS)
    size = (int(cam.get(cv2.CAP_PROP_FRAME_WIDTH)), int(cam.get(cv2.CAP_PROP_FRAME_HEIGHT)))
    while True:
        ret, frame = cam.read()
        # print ()
        if ret is True:
            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        else:
            print ('can not read the frame \n')
            break
        faces = face_xml.detectMultiScale(gray, 1.3, 5)
        print ('face=', len(faces))
        #框选检测到的人脸,获得人脸所在区域
        for (x, y, w, h) in faces:
            cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
            roi_face = gray[y:y + h, x:x + w]
            roi = frame[y:y + h, x:x + w]
            eyes = eye_xml.detectMultiScale(roi_face)
            for (x_, y_, w_, h_) in eyes:
                cv2.rectangle(roi, (x_, y_), (x_ + w_, y_ + h_), (0, 255, 125), 2)
        cv2.imshow('image', frame)

        ch = cv2.waitKey(1000/int(fps))
        if ch == 27:
            break
    cv2.destroyAllWindows()