C++版本OpenCV实现对mnist手写数据集的解析

在机器学习中，mnist数据集是一个使用很广泛的数据集，一般教程中都是python对mnist数据集的解析，在本文中将采用C++语言进行手写数据集的数据提取。
环境准备：
vs2015
OpenCV4.5.0

首先在mnist官网上下载mnist数据集的压缩包。

下载完成后将压缩包解压得到idx3-ubyte格式文件，如下图所示：

下面代码为C++版解析代码：

#include<iostream>
#include<opencv.hpp>
#include <string>
#include <fstream>
using namespace std;
using namespace cv;


//小端存储转换
int reverseInt(int i);
//读取image数据集信息
Mat read_mnist_image(const string fileName);
//读取label数据集信息
Mat read_mnist_label(const string fileName);

string train_images_path = "G:/vs2015_opencv_ml/mnist/train-images.idx3-ubyte";
string train_labels_path = "G:/vs2015_opencv_ml/mnist/train-labels.idx1-ubyte";

int main()
{
	//读取标签数据 (60000,1) 类型为int32
	Mat train_labels = read_mnist_label(train_labels_path);
	//打印第一个标签
	Mat label0 = train_labels.rowRange(0, 1);
	cout << "第一个标签:" << label0 << endl;


	//读取图像数据 (60000,784) 类型为float32 数据未归一化
	Mat train_images = read_mnist_image(train_images_path);

	//取出第0张图像
	Mat img0 = train_images.rowRange(0, 1);//(1,784)
	//改变形状为(28,28)
	img0 = img0.reshape(1, 28);
	//显示图像
	cv::imshow("test", img0);
	cv::waitKey(0);

	getchar();
	return 0;
}


;

int reverseInt(int i) {
	unsigned char c1, c2, c3, c4;

	c1 = i & 255;
	c2 = (i >> 8) & 255;
	c3 = (i >> 16) & 255;
	c4 = (i >> 24) & 255;

	return ((int)c1 << 24) + ((int)c2 << 16) + ((int)c3 << 8) + c4;
}

Mat read_mnist_image(const string fileName) {
	int magic_number = 0;
	int number_of_images = 0;
	int n_rows = 0;
	int n_cols = 0;

	Mat DataMat;

	ifstream file(fileName, ios::binary);
	if (file.is_open())
	{
		cout << "成功打开图像集 ..." << endl;

		file.read((char*)&magic_number, sizeof(magic_number));//幻数（文件格式）
		file.read((char*)&number_of_images, sizeof(number_of_images));//图像总数
		file.read((char*)&n_rows, sizeof(n_rows));//每个图像的行数
		file.read((char*)&n_cols, sizeof(n_cols));//每个图像的列数

		magic_number = reverseInt(magic_number);
		number_of_images = reverseInt(number_of_images);
		n_rows = reverseInt(n_rows);
		n_cols = reverseInt(n_cols);
		cout << "幻数（文件格式）:" << magic_number
			<< " 图像总数:" << number_of_images
			<< " 每个图像的行数:" << n_rows
			<< " 每个图像的列数:" << n_cols << endl;

		cout << "开始读取Image数据......" << endl;

		DataMat = Mat::zeros(number_of_images, n_rows * n_cols, CV_32FC1);
		for (int i = 0; i < number_of_images; i++) {
			for (int j = 0; j < n_rows * n_cols; j++) {
				unsigned char temp = 0;
				file.read((char*)&temp, sizeof(temp));
				//可以在下面这一步将每个像素值归一化
				float pixel_value = float(temp);
				//按照行将像素值一个个写入Mat中
				DataMat.at<float>(i, j) = pixel_value;
			}
		}

		cout << "读取Image数据完毕......" << endl;

	}
	file.close();
	return DataMat;
}

Mat read_mnist_label(const string fileName) {
	int magic_number;
	int number_of_items;

	Mat LabelMat;

	ifstream file(fileName, ios::binary);
	if (file.is_open())
	{
		cout << "成功打开标签集 ... " << endl;

		file.read((char*)&magic_number, sizeof(magic_number));
		file.read((char*)&number_of_items, sizeof(number_of_items));
		magic_number = reverseInt(magic_number);
		number_of_items = reverseInt(number_of_items);

		cout << "幻数（文件格式）:" << magic_number << "  ;标签总数:" << number_of_items << endl;

		cout << "开始读取Label数据......" << endl;
		//CV_32SC1代表32位有符号整型 通道数为1
		LabelMat = Mat::zeros(number_of_items, 1, CV_32SC1);
		for (int i = 0; i < number_of_items; i++) {
			unsigned char temp = 0;
			file.read((char*)&temp, sizeof(temp));
			LabelMat.at<unsigned int>(i, 0) = (unsigned int)temp;
		}
		cout << "读取Label数据完毕......"<<endl;

	}
	file.close();
	return LabelMat;
}

运行程序，执行结果如下：