deepid人脸识别

#include "FaceRecognition.h" #include <facedetect-dll.h> #include <Windows.h> #include <sstream> #pragma comment(lib,"libfacedetect-x64.lib") #define DETECT_BUFFER_SIZE 0x20000 namespace caffe { extern INSTANTIATE_CLASS(InputLayer); extern INSTANTIATE_CLASS(InnerProductLayer); extern INSTANTIATE_CLASS(DropoutLayer); extern INSTANTIATE_CLASS(ConvolutionLayer); REGISTER_LAYER_CLASS(Convolution); extern INSTANTIATE_CLASS(ReLULayer); REGISTER_LAYER_CLASS(ReLU); extern INSTANTIATE_CLASS(PoolingLayer); REGISTER_LAYER_CLASS(Pooling); extern INSTANTIATE_CLASS(LRNLayer); REGISTER_LAYER_CLASS(LRN); extern INSTANTIATE_CLASS(SoftmaxLayer); REGISTER_LAYER_CLASS(Softmax); extern INSTANTIATE_CLASS(ConcatLayer); //REGISTER_LAYER_CLASS(Concat); extern INSTANTIATE_CLASS(MemoryDataLayer); extern INSTANTIATE_CLASS(EltwiseLayer); //REGISTER_LAYER_CLASS(Eltwise); extern INSTANTIATE_CLASS(PowerLayer); //REGISTER_LAYER_CLASS(Power); extern INSTANTIATE_CLASS(SplitLayer); //REGISTER_LAYER_CLASS(); extern INSTANTIATE_CLASS(BatchNormLayer); //REGISTER_LAYER_CLASS(BatchNorm); extern INSTANTIATE_CLASS(ScaleLayer); extern INSTANTIATE_CLASS(BiasLayer); extern INSTANTIATE_CLASS(PReLULayer); extern INSTANTIATE_CLASS(AccuracyLayer); extern INSTANTIATE_CLASS(SoftmaxWithLossLayer); } FaceRecognition::FaceRecognition() { net = new caffe::Net<float>("deepId_deploy.prototxt", caffe::TEST); net->CopyTrainedLayersFrom("_iter_70000.caffemodel"); memory_layer = (caffe::MemoryDataLayer<float> *)net->layers()[0].get(); } FaceRecognition::~FaceRecognition() { } bool FaceRecognition::generateFaceBase(std::vector<std::string> files) { //人脸库创建来源是硬盘上的单照文件，所以不需要对照片进行缩放处理 //标签名即存放为文件名 cv::Rect roi;//创建感兴趣区域，便于进行图像切割和处理 for (int i = 0; i < files.size(); i++) { //对人脸库的每一张人脸照片进行特征扫描并记录 cv::Mat frame; frame = cv::imread(files[i].c_str()); if (detectSingleFace(frame,roi)) { //如果检测到人脸，此处只会有一张人脸 cv::Mat img_224 = frame(roi);//根据得到的感兴趣区域，从原图中截取出来，并保存成224*224的图片 //cv::cvtColor(img_224, img_224, CV_BGR2GRAY); resize(img_224, img_224, cv::Size(224, 224)); //获取人脸库标签（以文件名为标签） int pos = files[i].find_last_of('\\'); std::string label = files[i].substr(pos + 1); //存储进入人脸库 SingleFace tempface; tempface.sourceImage = frame; tempface.position = roi; tempface.Roi_224 = img_224; auto feature_ = extractFeature(tempface.Roi_224); tempface.feature = feature_; tempface.label = label; FaceBase.push_back(tempface); std::thread t2(&FaceRecognition::StoreFace, this, i); t2.detach(); } else { printf("no base face detected!!\n"); return false; } } return true; } bool FaceRecognition::generateFaceArray(cv::Mat input) { int scale;//图片检索前图像缩放比例 cv::Mat smallImg; //存放缩放只后的待检索图片 std::vector<cv::Rect> rois; //对图片检索后存储所有感兴趣区域（所有人脸） scale = imgScale(input, smallImg); //printf("scale=%d", scale); if (detectMultiFace(smallImg,scale,rois)) { m_nFaceArraySize = rois.size(); for (int i = 0; i < rois.size(); i++) { //对于每一张检测出来的脸，都在原图上进行图像切割，并进行大小重定义 cv::Mat img_224 = input(rois[i]);//根据得到的感兴趣区域，从原图中截取出来，并保存成224*224的图片 //cv::cvtColor(img_224, img_224, CV_BGR2GRAY); cv::resize(img_224, img_224, cv::Size(55, 55)); //存储进入检索人脸组 SingleFace tempface; tempface.sourceImage = input; tempface.position = rois[i]; tempface.Roi_224 = img_224; //auto feature_ = extractFeature(tempface.Roi_224); //tempface.feature = feature_; FaceArray.push_back(tempface); std::thread t1(&FaceRecognition::StoreFaces, this, i); t1.detach(); //rectangle(smallImg, cv::Rect((rois[i].x / scale), (rois[i].y / scale), (rois[i].width / scale), (rois[i].height / scale)), cv::Scalar(0, 255, 0)); //imshow("缩放后图片", smallImg); //imshow(s, img_224); //cv::waitKey(); } } else { printf("no face detected!!!!!!\n"); return false; } return true; } //多线程启动，用于将检索人脸信息写入人脸集合 void FaceRecognition::StoreFaces(int index) { //互斥写入人脸集合 thread_mutex1.lock(); //auto feature = ExtractFeature(m_vecFaceArray[index].Roi_224); FaceArray[index].feature = extractFeature(FaceArray[index].Roi_224); m_nFaceArraySize--; thread_mutex1.unlock(); } //多线程启动，用于将带标签人脸库信息写入人脸集合 void FaceRecognition::StoreFace(int index) { //互斥写入人脸库 thread_mutex2.lock(); //auto feature = extractFeature(FaceBase[index].Roi_224); FaceBase[index].feature = extractFeature(FaceBase[index].Roi_224); m_nFaceBaseSize--; thread_mutex2.unlock(); } std::vector<float> FaceRecognition::extractFeature(cv::Mat face_224) { thread_mutex3.lock(); std::vector<cv::Mat> test{ face_224 }; std::vector<int> testLabel{ 0 }; memory_layer->AddMatVector(test, testLabel);// memory_layer and net , must be define be a global variable. std::vector<caffe::Blob<float>*> input_vec; net->Forward(input_vec); auto fc7 = net->blob_by_name("fc160");//提取fc7层！4096维特征 float* begin = fc7->mutable_cpu_data(); std::vector<float> feature{ begin, begin + fc7->channels() }; //cout << fc7->channels(); thread_mutex3.unlock(); return move(feature); } SingleFace FaceRecognition::recognizeFaces(SingleFace singleface) { //输入人脸库照片，与人脸检索照片对比，得到结果 int size_ = FaceArray.size();//获取人脸检索的数量 //获取单张人脸库的特征向量位置 float *single_feature = &singleface.feature[0]; //得到特征向量的维度 int single_channel = singleface.feature.size(); std::vector<double> like_array;//计算检索人脸与人脸库每张照片的相似度,取相似度最高的一个 for (int i = 0; i < size_; i++) { float *faces_feature = &FaceArray[i].feature[0];//人脸库里面的特征 like_array.push_back(likeValue(single_feature, faces_feature, single_channel)); } std::vector<double>::iterator biggest = std::max_element(std::begin(like_array), std::end(like_array)); int max_ = distance(std::begin(like_array), biggest); printf("匹配度为%f\n", *biggest); return FaceArray[max_];//返回匹配的特征脸 } bool FaceRecognition::detectSingleFace(cv::Mat input, cv::Rect &roi) { cv::Mat gray; cv::cvtColor(input, gray, CV_BGR2GRAY);//转成灰度图 int * pResults = NULL; unsigned char * pBuffer = (unsigned char *)malloc(DETECT_BUFFER_SIZE);//分配缓冲计算空间 if (!pBuffer) { fprintf(stderr, "Can not alloc buffer.\n"); return false; } int doLandmark = 1; //特征点检测模式 pResults = facedetect_multiview_reinforce(pBuffer, (unsigned char*)(gray.ptr(0)), gray.cols, gray.rows, (int)gray.step, 1.05f, 3, 2, 0, doLandmark); //printf("%d faces detected.\n", (pResults ? *pResults : 0));//显示检测到的人脸数量（人脸库为1） //需要获取人脸大小及位置，不需要对特征进行标记 int face_num = (pResults ? *pResults : 0); if (face_num == 0) { printf("no face detected"); return false; } cv::Mat result_multiview_reinforce = input.clone();//只用来显示人脸图像而已 //获取人脸位置坐标 short * p = ((short*)(pResults + 1)); int x = p[0]; int y = p[1]; int w = p[2]; int h = p[3]; roi = cv::Rect(x, y, w, h); //rectangle(result_multiview_reinforce, cv::Rect(x, y, w, h), cv::Scalar(0, 255, 0)); //imshow("result_multiview_reinforce", result_multiview_reinforce); //cv::waitKey(); free(pBuffer);//释放空间，每次都重新分配 return true; } bool FaceRecognition::detectMultiFace(cv::Mat input, int scale, std::vector<cv::Rect> &rois) { cv::Mat gray; cv::cvtColor(input, gray, CV_BGR2GRAY);//转成灰度图 int * pResults = NULL; unsigned char * pBuffer = (unsigned char *)malloc(DETECT_BUFFER_SIZE);//分配缓冲计算空间 if (!pBuffer) { fpri