#eigenface

PCA and DCT Based Approach for Face Recognition

by Manish Varyani | Pallavi Narware | Lokendra Singh Banafar "PCA and DCT Based Approach for Face Recognition" 

Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-3 | Issue-3 , April 2019, 

URL: https://www.ijtsrd.com/papers/ijtsrd23283.pdf

Paper URL: https://www.ijtsrd.com/engineering/electronics-and-communication-engineering/23283/pca-and-dct-based-approach-for-face-recognition/manish-varyani

international journals in engineering, engineering journal, paper publication for engineering

The word “eigenface”, half German and half English, can be translated roughly as “one’s own face”. In terms of a technical definition, an eigenface is a two-dimensional, grayscale image representing distinctive characteristics of a facial image.

The distinctive features are stored by the system as coefficients.Advocates of this technology maintain that most human faces can be reconstructed by combining features drawn from a set of between 100 and 150 such eigenfaces. In other words, although there are more than 5 billion individual human faces on this planet, they can all be reconstructed from fewer than 150 eigenfaces. Eigenface components are used to build facialimages. Software applications using this technology analyze a facial image, calculate its eigenface value using distinctive feature coefficients (essentially deconstructing the face into its eigenface components), and compare that value to a biometric template or database of biometrictemplates to find matching or similar eigenface values.

OpenCV2.4.1から実装されている顔認識手法です．

ざっりくいうと，多くの顔画像から重要な成分のみを取り出して識別に使用するイメージです．

※識別に使う時に共通する成分（平均顔）を原画像から引いてから，識別すると性能がいいと報告があります．

http://docs.opencv.org/modules/contrib/doc/facerec/facerec_tutorial.html

各ソースコードとスクリプトの場所

C:\OpenCV2.4.2\modules\contrib\doc\facerec\src

上の2つの情報が十分にあるので，ここではシンプルにしたコードを紹介します．

流れ

①画像の読み込み

②ラベル付け

③trainと書き出し

④読み込み

⑤識別したい画像が顔画像かどうかの簡単なチェック

⑥predictによる識別

※学習と識別画像のサイｚは統一する必要があります．

このプログラムの現状は２クラス分類です．

例えば，AとＢさんの写真を用意する.

Aさんには１

Ｂさんには-１のラベルつける

新しい写真をもってきて，1 -1を判定する．

1ならＡさん

-1ならＢさんと判断されます．

ラベルをx種類つければ，x分類されてかえってくると思います．

ちなみに⑤をいれたのは，明らかに顔以外の画像にでも，1か－１を返してくるので，前処理として入れました．

-100なら，顔じゃないと判断して計算打ち切るのがいいと思います．

以下，ソースコード

static Mat norm_0_255(InputArray _src) {

    Mat src = _src.getMat();

    // Create and return normalized image:

    Mat dst;

    switch(src.channels()) {

    case 1:

        cv::normalize(_src, dst, 0, 255, NORM_MINMAX, CV_8UC1);

        break;

    case 3:

        cv::normalize(_src, dst, 0, 255, NORM_MINMAX, CV_8UC3);

        break;

    default:

        src.copyTo(dst);

        break;

    }

    return dst;

}

// Compare two images by getting the L2 error (square-root of sum of squared error).

double getSimilarity(const Mat& A, const Mat& B)

{

    if (A.rows > 0 && A.rows == B.rows && A.cols > 0 && A.cols == B.cols) {

        // Calculate the L2 relative error between the 2 images.

        double errorL2 = norm(A, B, CV_L2);

        // Convert to a reasonable scale, since L2 error is summed across all pixels of the image.

        double similarity = errorL2 / (double)(A.rows * A.cols);

if(similarity>THRESH_HOLD)

        return 100;

else

return -100;

    }

    else {

        //cout << "WARNING: Images have a different size in 'getSimilarity()'." << endl;

        return -999999;  // Return a bad value

    }

}

以下　main文

int ALL=8;//写真の枚数

std::stringstream ofs;

 vector<Mat> images;

 vector<int> labels;

 //////////////////

 for(int i=0;i<ALL;i++){

 ofs.str("");

 ofs<<"IMG/"<<i<<".jpg";

 cv::Mat temp =cv::imread(ofs.str(),0);

   cv::Mat resize(HEIGHT,WIDTH, CV_8U);

cv::resize(temp,resize,resize.size());

 images.push_back(resize);

 }

///////

 for(int i=0;i<ALL;i++){

 if(i<MAN)

 labels.push_back(1);

 else

 labels.push_back(-1);

 }

Ptr<FaceRecognizer> learn = createEigenFaceRecognizer();//学習して書き出す．

Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//読み込んで識別に使用

learn->train(images, labels);

learn->save("pca.xml");

model->load("pca.xml");

// Here is how to get the eigenvalues of this Eigenfaces model:

    Mat eigenvalues = model->getMat("eigenvalues");

    // And we can do the same to display the Eigenvectors (read Eigenfaces):

    Mat W = model->getMat("eigenvectors");

    // Get the sample mean from the training data

    Mat mean = model->getMat("mean");

    // Display or save:

         imshow("平均顔(mean)", norm_0_255(mean.reshape(1, images[0].rows)));

cv::waitKey(0);

cv::Mat temp = cv::imread("IMG/image_0014.jpg",0);

cv::Mat resize(256,256, CV_8U);

cv::resize(temp,resize,resize.size());

cv::Mat pre_temp=cv::imread("IMG/0.jpg",0);

cv::Mat pre_reisze_face(HEIGHT,WIDTH,CV_8U);

cv::resize(pre_temp,pre_reisze_face,pre_reisze_face.size());

std::cout<<getSimilarity(resize,pre_reisze_face)<<"\n";

The (literally) average American Elf strip - its "eigencomic". First image is the average of all 5106 strips, second image shows 80 strips per sample, from 1998 in the upper left corner to 2013 at the bottom right.