tensorflow开发 之 mnist文档解析

  mnist本身是将数字手写体图像，归一化后信息合成的文档，分训练集和测试集两部分，每部分包含图片和标签两个文档，本文使用C++写的Demo将文档解码，并通过opencv将数据存储为常规的图像文档。源码下载地址：https://github.com/zacSuo/mnist

• t10k-images.idx3-ubyte
• t10k-labels.idx1-ubyte
• train-images.idx3-ubyte
• train-labels.idx1-ubyte

使用说明

环境要求

• visual studio 2012及以上版本
• opencv 2.4及以上版本

  配置

  启动前先通过项目属性，设置本地opencv的路径。
  

执行结果

  在目录的两个文档夹，分别将测试集（test_images）和训练集（train_images）的图片保存到其中。

进程说明

解码图像

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

int (std::string filename, std::vector<cv::Mat> &vec) { int number_of_images = 0; std::ifstream file(filename, std::ios::binary); if (file.is_open()) { int magic_number = 0; int n_rows = 0; int n_cols = 0; file.read((char*)&magic_number, sizeof(magic_number)); magic_number = ReverseInt(magic_number); file.read((char*)&number_of_images, sizeof(number_of_images)); number_of_images = ReverseInt(number_of_images); file.read((char*)&n_rows, sizeof(n_rows)); n_rows = ReverseInt(n_rows); file.read((char*)&n_cols, sizeof(n_cols)); n_cols = ReverseInt(n_cols); for (int i = 0; i < number_of_images; ++i) { cv::Mat tp = cv::Mat::zeros(n_rows, n_cols, CV_8UC1); for (int r = 0; r < n_rows; ++r) { for (int c = 0; c < n_cols; ++c) { unsigned char temp = 0; file.read((char*)&temp, sizeof(temp)); tp.at<uchar>(r, c) = (int)temp; } } vec.push_back(tp); } file.close(); } return number_of_images; }

解码图像对应数值

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

void read_Mnist_Label(std::string filename, std::vector<int> &vec) { std::ifstream file(filename, std::ios::binary); if (file.is_open()) { int magic_number = 0; int number_of_images = 0; int n_rows = 0; int n_cols = 0; file.read((char*)&magic_number, sizeof(magic_number)); magic_number = ReverseInt(magic_number); file.read((char*)&number_of_images, sizeof(number_of_images)); number_of_images = ReverseInt(number_of_images); for (int i = 0; i < number_of_images; ++i) { unsigned char temp = 0; file.read((char*)&temp, sizeof(temp)); vec[i] = (int)temp; } file.close(); } }

生成图像文档

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

int decodeDatabase(std::string filenameImages,std::string filenameLabels,std::string imagesPathSave) { std::vector<cv::Mat> vec_train_images; int number_of_train_images = read_Mnist(filenameImages, vec_train_images); std::vector<int> vec_train_labels(number_of_train_images); read_Mnist_Label(filenameLabels, vec_train_labels); if (vec_train_images.size() != vec_train_labels.size()) { std::cout << "parse MNIST train file error" << std::endl; return -1; } // save images int count_digits[10]; std::fill(&count_digits[0], &count_digits[0] + 10, 0); for (int i = 0; i < vec_train_images.size(); i++) { int number = vec_train_labels[i]; std::string image_name = GetImageName(number, count_digits); image_name = imagesPathSave + image_name + ".jpg"; cv::imwrite(image_name, vec_train_images[i]); } return 0; }

每个数值各选20张图像合并成一个图像文档

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

int mergeSomeImages() { std::string images_path = ".././mnistFile/train_images/"; int width = 28 * 20; int height = 28 * 10; cv::Mat dst(height, width, CV_8UC1); for (int i = 0; i < 10; i++) { for (int j = 1; j <= 20; j++) { int x = (j-1) * 28; int y = i * 28; cv::Mat part = dst(cv::Rect(x, y, 28, 28)); std::string str = std::to_string(j); if (j < 10) str = "0000" + str; else str = "000" + str; str = std::to_string(i) + "_" + str + ".jpg"; std::string input_image = images_path + str; cv::Mat src = cv::imread(input_image, 0); if (src.empty()) { fprintf(stderr, "read image error: %sn", input_image.c_str()); return -1; } src.copyTo(part); } } std::string output_image = images_path + "result.png"; cv::imwrite(output_image, dst); return 0; }

原文链接 大专栏  https://www.dazhuanlan.com/2019/08/25/5d61bcc0916e4/