「何謂OpenCV?」
Open代表Open Source，CV為Computer Vision，Computer Vision被廣泛應用在計算機科學及醫學等領域，OpenCV是由Intel公司參與及開發出來的，感謝Intel的Open Source :)。OpenCV支援Windows跟Linux兩種平台，並具有強大的圖形跟矩陣的運算。然而，近幾年OpenCV為人垢病的是，它並不再更新，因此無法提供一些新的演算法。But，這個問題似乎得到了回應，最近OpenCV已經開始在改版了，加入了一些新的函數，像是近年來在電腦視覺領域上，極常被使用及提及的Scale-invariant feature transform(SIFT)。
「下載及安裝OpenCV」
最新版的OpenCV可至以下網址下載:
http://sourceforge.net/projects/opencvlibrary/
「Face Detection」
There are numerous researches making efforts toward how to correctly detect faces in digital images. 而OpenCV所使用的方法是由「Viola & Jones」所發表的AdaBoost Learning with Haar-Like Features來實現人臉偵測。下面就簡單介紹一下這個方法:- 首先定義一些Haar-Like Features。
- 再來，我們必須給定一些sample，例如: 假如我們要偵測的是人臉，那麼我們就要輸入一些人臉的sample。有了這些sample，我們將會利用AdaBoost learning algorithm來挑出某幾個代表性的Haar-Like Features，以這個例子來講，我們可以想成這些被挑選出來的feature就是代表人臉的feature。
- 每個被挑選出來的feature都代表一種classifier，許多種被挑選出來的feature因此構成了一連串的classifier，我們稱為strong classifier。而每個classifier皆用來判斷所輸入的圖片是否為人臉，並且回傳「是」或「否」，最後，通過所有classifier的圖片將被判定是一張人臉。
以上就是由Viola與Jones所提出的Object Detection方法，除了人臉，當然它還可以拿來用在其它的物件偵測上，只要我們改變一開始輸入訓練用的sample型態即可。
「Source Code」
OpenCV 有提供Face Detection的source code，不過它註解很少，並且有點難懂，所以我試著修改它，並加上一些註解/*
* Face Detection in Digital Photographs
*
* We leverage the AdaBoost learning-based face detection method with Haar-like features to detect faces in photos.
* However, OpenCV supplies Haar Cascade classifier for object (face) detection.
*
* (C) Matt Swanson & TrekLee
* http://seeyababy.blogspot.com/
*
*/
// Path setting for OpenCV Library
#pragma comment(lib,"../OpenCV/lib/cv.lib")
#pragma comment(lib,"../OpenCV/lib/cxcore.lib")
#pragma comment(lib,"../OpenCV/lib/highgui.lib")
// Path setting for OpenCV Header files
#include "../OpenCV/include/cv.h"
#include "../OpenCV/include/cxcore.h"
#include "../OpenCV/include/highgui.h"
#include <stdio.h>
#include <iostream>
using namespace std;
// the minimum object size
int min_face_height = 20;
int min_face_width = 10;
//Face Detection
int runFaceDetection(IplImage* image_detect);
void main()
{
IplImage* pImg = cvLoadImage( "image.jpg", 1);
runFaceDetection(pImg);
cvReleaseImage( &pImg );
}
int runFaceDetection(IplImage* image_detect)
{
// Load the pre-trained Haar classifier data.
CvHaarClassifierCascade* classifier = (CvHaarClassifierCascade*)cvLoad("haarcascade_frontalface_alt.xml", 0, 0, 0);
// Quit the application if the input source or the classifier data failed to load properly.
if( !classifier)
{
cerr << "ERROR: Could not load classifier cascade." << endl;
return -1;
}
// Create a CvMemStorage object for use by the face detection function.
CvMemStorage* facesMemStorage = cvCreateMemStorage(0);
IplImage* tempFrame = cvCreateImage(cvSize(image_detect->width,
image_detect->height), IPL_DEPTH_8U, image_detect->nChannels);
// Copy the current frame into the temporary image. Also, make
// sure the images have the same orientation.
if(image_detect->origin == IPL_ORIGIN_TL)
{
cvCopy(image_detect, tempFrame, 0);
}
else
{
cvFlip(image_detect, tempFrame, 0);
}
/* Perform face detection on the temporary image, adding a rectangle around the detected face. */
// "Resets" the memory but does not deallocate it.
cvClearMemStorage(facesMemStorage);
// Run the main object recognition function. The arguments are:
// 1. the image to use
// 2. the pre-trained Haar classifier cascade data
// 3. memory storage for rectangles around recognized objects
// 4. a scale factor "by which the search window is scaled between the
// subsequent scans, for example, 1.1 means increasing window by 10%"
// 5. the "minimum number (minus 1) of neighbor rectangles that makes up
// an object. All the groups of a smaller number of rectangles than
// min_neighbors-1 are rejected. If min_neighbors is 0, the function
// does not any grouping at all and returns all the detected candidate
// rectangles, which may be useful if the user wants to apply a
// customized grouping procedure."
// 6. flags which determine the mode of operation
// 7. the minimum object size (if possible, increasing this will
// really speed up the process)
CvSeq* faces = cvHaarDetectObjects(tempFrame, classifier, facesMemStorage, 1.1,
2, CV_HAAR_DO_CANNY_PRUNING, cvSize(min_face_width, min_face_height));
// If any faces were detected, draw rectangles around them.
if (faces)
{
for(int i = 0; i < faces->total; ++i)
{
// Setup two points that define the extremes of the rectangle,
// then draw it to the image..
CvPoint point1, point2;
CvRect* rectangle = (CvRect*)cvGetSeqElem(faces, i);
point1.x = rectangle->x;
point2.x = rectangle->x + rectangle->width;
point1.y = rectangle->y;
point2.y = rectangle->y + rectangle->height;
cvRectangle(tempFrame, point1, point2, CV_RGB(255,0,0), 3, 8, 0);
}
}
// Show the result in the window.
cvNamedWindow("Face Detection Result", 1);
cvShowImage("Face Detection Result", tempFrame);
cvWaitKey(0);
cvDestroyWindow("Face Detection Result");
// Clean up allocated OpenCV objects.
cvReleaseMemStorage(&facesMemStorage);
cvReleaseImage(&tempFrame);
cvReleaseHaarClassifierCascade( &classifier );
return 0;
}「Experiment」
接下來我挑張照片來做個簡單的實驗，下圖是做完人臉偵測，並框出系統判斷為人臉的結果
我們可以發現接近右下角的地方有個誤判的情況發生，但不得不說的是，仔細看它真的長的很像一張邪惡的人臉...囧|||...，回歸正題，我們該如何解決這個誤判? 除了加入膚色判斷，或是讓face detector更準確之外，其實有個立即就可以針對這個問題來解決的方法，當我們觀察以下這個函式
CvSeq* faces = cvHaarDetectObjects(tempFrame, classifier, facesMemStorage, 1.1, 2, CV_HAAR_DO_CANNY_PRUNING, cvSize(min_face_width, min_face_height));
其中，cvSize(min_face_width, min_face_height)設定被偵測人臉的最小size，上面被誤判為人臉的圖片明顯小於其它正常的人臉，所以我們可以考慮將size放大，看看是否可以解決這個問題，因此我將min_face_height及min_face_width由原本的size各加上10個pixel，重新偵測所得到的結果如下
我們發現誤判果然被解決了。
我們發現誤判果然被解決了。
「Conclusion」
本篇文章首先介紹了由Viola與Jones所提出的利用AdaBoost Learning with Haar-Like Features來做人臉偵測的方法，並且提供了source code，這份程式改進了OpenCV所提供的source code其難讀(for me)、註解少的缺失，最後，做了一個簡單的實驗及討論。
Coming soon...
