考虑到一些朋友对鄙人的性别判断方法感兴趣,而自己的代码又没妥善保存,因此特在这里进行弥补,将作业报告PPT内容公布,并进行简单地代码说明,希望对感兴趣的童鞋能有所帮助。
代码:
代码主要根据opencv\modules\contrib\doc\facerec\src\facerec_fisherfaces.cpp来讲解,源代码内容如下:
/* * Copyright (c) 2011. Philipp Wagner <bytefish[at]gmx[dot]de>. * Released to public domain under terms of the BSD Simplified license. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * Neither the name of the organization nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * See <http://www.opensource.org/licenses/bsd-license> */ #include "opencv2/core/core.hpp" #include "opencv2/contrib/contrib.hpp" #include "opencv2/highgui/highgui.hpp" #include <iostream> #include <fstream> #include <sstream> using namespace cv; using namespace std; 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; } static void read_csv(const string& filename, vector<Mat>& images, vector<int>& labels, char separator = ';') { std::ifstream file(filename.c_str(), ifstream::in); if (!file) { string error_message = "No valid input file was given, please check the given filename."; CV_Error(CV_StsBadArg, error_message); } string line, path, classlabel; while (getline(file, line)) { stringstream liness(line); getline(liness, path, separator); getline(liness, classlabel); if(!path.empty() && !classlabel.empty()) { images.push_back(imread(path, 0)); labels.push_back(atoi(classlabel.c_str())); } } } int main(int argc, const char *argv[]) { // Check for valid command line arguments, print usage // if no arguments were given. if (argc < 2) { cout << "usage: " << argv[0] << " <csv.ext> <output_folder> " << endl; exit(1); } string output_folder; if (argc == 3) { output_folder = string(argv[2]); } // Get the path to your CSV. string fn_csv = string(argv[1]); // These vectors hold the images and corresponding labels. vector<Mat> images; vector<int> labels; // Read in the data. This can fail if no valid // input filename is given. try { read_csv(fn_csv, images, labels); } catch (cv::Exception& e) { cerr << "Error opening file \"" << fn_csv << "\". Reason: " << e.msg << endl; // nothing more we can do exit(1); } // Quit if there are not enough images for this demo. if(images.size() <= 1) { string error_message = "This demo needs at least 2 images to work. Please add more images to your data set!"; CV_Error(CV_StsError, error_message); } // Get the height from the first image. We'll need this // later in code to reshape the images to their original // size: int height = images[0].rows; // The following lines simply get the last images from // your dataset and remove it from the vector. This is // done, so that the training data (which we learn the // cv::FaceRecognizer on) and the test data we test // the model with, do not overlap. Mat testSample = images[images.size() - 1]; int testLabel = labels[labels.size() - 1]; images.pop_back(); labels.pop_back(); // The following lines create an Fisherfaces model for // face recognition and train it with the images and // labels read from the given CSV file. // If you just want to keep 10 Fisherfaces, then call // the factory method like this: // // cv::createFisherFaceRecognizer(10); // // However it is not useful to discard Fisherfaces! Please // always try to use _all_ available Fisherfaces for // classification. // // If you want to create a FaceRecognizer with a // confidence threshold (e.g. 123.0) and use _all_ // Fisherfaces, then call it with: // // cv::createFisherFaceRecognizer(0, 123.0); // Ptr<FaceRecognizer> model = createFisherFaceRecognizer(); model->train(images, labels); // The following line predicts the label of a given // test image: int predictedLabel = model->predict(testSample); // // To get the confidence of a prediction call the model with: // // int predictedLabel = -1; // double confidence = 0.0; // model->predict(testSample, predictedLabel, confidence); // string result_message = format("Predicted class = %d / Actual class = %d.", predictedLabel, testLabel); cout << result_message << endl; // 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: if(argc == 2) { imshow("mean", norm_0_255(mean.reshape(1, images[0].rows))); } else { imwrite(format("%s/mean.png", output_folder.c_str()), norm_0_255(mean.reshape(1, images[0].rows))); } // Display or save the first, at most 16 Fisherfaces: for (int i = 0; i < min(16, W.cols); i++) { string msg = format("Eigenvalue #%d = %.5f", i, eigenvalues.at<double>(i)); cout << msg << endl; // get eigenvector #i Mat ev = W.col(i).clone(); // Reshape to original size & normalize to [0...255] for imshow. Mat grayscale = norm_0_255(ev.reshape(1, height)); // Show the image & apply a Bone colormap for better sensing. Mat cgrayscale; applyColorMap(grayscale, cgrayscale, COLORMAP_BONE); // Display or save: if(argc == 2) { imshow(format("fisherface_%d", i), cgrayscale); } else { imwrite(format("%s/fisherface_%d.png", output_folder.c_str(), i), norm_0_255(cgrayscale)); } } // Display or save the image reconstruction at some predefined steps: for(int num_component = 0; num_component < min(16, W.cols); num_component++) { // Slice the Fisherface from the model: Mat ev = W.col(num_component); Mat projection = subspaceProject(ev, mean, images[0].reshape(1,1)); Mat reconstruction = subspaceReconstruct(ev, mean, projection); // Normalize the result: reconstruction = norm_0_255(reconstruction.reshape(1, images[0].rows)); // Display or save: if(argc == 2) { imshow(format("fisherface_reconstruction_%d", num_component), reconstruction); } else { imwrite(format("%s/fisherface_reconstruction_%d.png", output_folder.c_str(), num_component), reconstruction); } } // Display if we are not writing to an output folder: if(argc == 2) { waitKey(0); } return 0; }
上面这段代码实现了用fisherface判断人脸性别的算法,但效果不好,因此我打算在fisherface之前插入一个pca降维过程,提高fisherface的精度。代码如下:
... // If you want to use _all_ Eigenfaces and have a threshold, // then call the method like this: // // cv::createEigenFaceRecognizer(0, 123.0); // Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型 model->train(images, labels);//训练pca模型,这里的model包含了所有特征值和特征向量,没有损失 model->save("eigenface.yml");//保存训练结果,供检测时使用 Mat eigenvalues = model->getMat("eigenvalues");//提取model中的特征值,该特征值默认由大到小排列 Mat W = model->getMat("eigenvectors");//提取model中的特征向量,特征向量的排列方式与特征值排列顺序一一对应 int xth = 121;//打算保留前121个特征向量,代码中没有体现原因,但选择121是经过斟酌的,首先,在我的实验中,"前121个特征值之和/所有特征值总和>0.97";其次,121=11^2,可以将结果表示成一个11*11的2维图像方阵,交给fisherface去计算。 vector<Mat> reduceDemensionimages;//降维后的图像矩阵 Mat evs = Mat(W, Range::all(), Range(0, xth));//选择前xth个特征向量,其余舍弃 for(int i=0;i<images.size();i++) { Mat projection = subspaceProject(evs, mean, images[i].reshape(1,1));//做子空间投影 reduceDemensionimages.push_back(projection.reshape(1,sqrt(xth));//将获得的子空间系数表示映射成2维图像,并保存起来 } Ptr<FaceRecognizer> fishermodel = createFisherFaceRecognizer(); fishermodel->train(reduceDemensionimages,labels);//用保存的降维后的图片来训练fishermodel,后面的内容与原始代码就没什么变化了 ...
这样就实现了pca+lda的训练过程,将训练好的pca和lda模型保存起来,在识别时用训练好的pca和lda模型对检测样本提取特征,根据其最终结果判断图片性别,就大功告成了。
pca+lda不但可以做性别识别,还可以做“年轻/年老”、“人种”、“表情”等多种识别,效果也不错,但这类识别都有一个特点:每类的样本个数大致相等,且分类应该是完备的,否则就不适合用pca+lda的方法来做,例如:在一大堆的图片中识别出梅西,这种事情就不太适合用pca+lda方法来做。
pca+lda方法在原理上和当前很火的深度学习存在着相似性,hinton大牛提出的多层结构:每层用无监督的方法提取特征,提取的特征又是下一层的输入,这一点和pca+lda很相似:pca是一种无监督的学习方法,提出的特征又作为lda方法的输入,大量实验证明多层深度学习会比浅层学习效果更好,这里的pca+lda就比单纯的lda或单纯的pca更好,大家也可以试试用多层pca+lda(即:pca+pca+...+pca+lda)会取的什么样的结果。