C语言解析MNIST数据集

布满荆棘的人生 2022-06-12 02:07 314阅读 0赞

### 小编用C语言解析MNIST数据集，产生的图片是倒立着的。 ###

**数据集下载地址：[http://yann.lecun.com/exdb/mnist/][http_yann.lecun.com_exdb_mnist]**  
**小编解析的是测试集：t10k-images.idx3-ubyte**  
**代码如下：**  
**bmp.h**

#ifndef BMP_H
    #define BMP_H
    
    #pragma pack(1)
    typedef struct tagBITMAPFILEHEADER
    {
        unsigned char bfType[2];//文件格式
        unsigned long bfSize;//文件大小
        unsigned short bfReserved1;//保留
        unsigned short bfReserved2;
        unsigned long bfOffBits; //DIB数据在文件中的偏移量
    }fileHeader;
    #pragma pack()
    /* 位图数据信息结构 */
    #pragma pack(1)
    typedef struct tagBITMAPINFOHEADER
    {
        unsigned long biSize;//该结构的大小
        long biWidth;//文件宽度
        long biHeight;//文件高度
        unsigned short biPlanes;//平面数
        unsigned short biBitCount;//颜色位数
        unsigned long biCompression;//压缩类型
        unsigned long biSizeImage;//DIB数据区大小
        long biXPixPerMeter;
        long biYPixPerMeter;
        unsigned long biClrUsed;//多少颜色索引表
        unsigned long biClrImporant;//多少重要颜色
    }fileInfo;
    #pragma pack()
    /* 调色板结构 */
    #pragma pack(1)
    typedef struct tagRGBQUAD
    {
        unsigned char rgbBlue; //蓝色分量亮度
        unsigned char rgbGreen;//绿色分量亮度
        unsigned char rgbRed;//红色分量亮度
        unsigned char rgbReserved;
    }rgbq;
    #pragma pack()
    
    #endif

**main**

#include <stdio.h>
    #include <stdlib.h>
    #include <iostream>
    #include "bmp.h"
    
    using namespace std;
    
    #define SUM 10000
    
    int main()
    {
        int a, num, size1, size2;
        //用于存储一张手写体图片的位图信息
        char image[28 * 28];
        //用于存储产生的图片的名字
        char name[100];
        //10000张测试集
        FILE *fp1 = fopen("C:\\Users\\Administrator\\Desktop\\data\\t10k-images.idx3-ubyte", "rb+");
        if (fp1 == NULL)
        {
            //cout << "unable open file1" << endl;
            //exit(1);
            printf("unable open file2");
        }
    
        /*
        //测试集的label
        FILE *fp11 = fopen("C:\\Users\\Administrator\\Desktop\\data\\t10k-labels.idx1-ubyte", "rb+");
        if (fp1 == NULL)
        {
            printf("unable open t10k-labels.idx1-ubyte");
        }
        else
        {
            //读取t10k-labels.idx1-ubyte的幻数、数量
            fread(&a, sizeof(int), 1, fp11);
            fread(&num, sizeof(int), 1, fp11);
            //读取t10k-labels.idx1-ubyte的前28×28个label
            fread(image, sizeof(char), 28 * 28, fp11);
            printf("------------------label--------------------------------");
            printf("a=%0x\n", a);
            printf("num=%0x\n", num);
            for (int i = 0; i < 28 * 28; i++)
            {
                printf("%d\t", image[i]);
            }
            printf("------------------label--------------------------------");
        }
        */
    
        //读取字符集的幻数、数量、宽、高
        fread(&a, sizeof(int), 1, fp1);
        fread(&num, sizeof(int), 1, fp1);
        fread(&size1, sizeof(int), 1, fp1);
        fread(&size2, sizeof(int), 1, fp1);
    
        /* //打印字符集的幻数、数量、宽、高 printf("a=%0x\n", a); printf("num=%0x\n", num); printf("size1=%0x\n", size1); printf("size2=%0x\n", size2); for (int i = 0; i < 28 * 28; i++) { printf("%d\t", image[i]); } */
    
        fileHeader fh;
        fileInfo fi;
        //编写信息头fi
        fi.biSize = 40;
        fi.biWidth = 28;
        fi.biHeight = 28;
        fi.biPlanes = 1;
        fi.biBitCount = 8;
        fi.biCompression = 0;
        fi.biSizeImage = 28 * 28;
        fi.biXPixPerMeter = 0;
        fi.biYPixPerMeter = 0;
        fi.biClrUsed = 0;
        fi.biClrImporant = 0;
    
        //编写文件头fh
        fh.bfType[0] = 0x42;
        fh.bfType[1] = 0x4D;
        fh.bfOffBits = sizeof(fileHeader) + sizeof(fileInfo) + 256 * sizeof(rgbq);
        fh.bfSize = fh.bfOffBits + fi.biSizeImage;
        fh.bfReserved1 = 0;
        fh.bfReserved2 = 0;
    
        //创建调色板
        rgbq *fq = (rgbq *)malloc(256 * sizeof(rgbq));
        for (int i = 0; i<256; i++)
        {
            fq[i].rgbBlue = fq[i].rgbGreen = fq[i].rgbRed = i;
        }
    
        //创建要生成的数字图片
        FILE *fp2 = fopen("C:\\Users\\Administrator\\Desktop\\data\\numbers.bmp", "wb");
        if (fp2 == NULL)
        {
            printf("unable open file2");
        }
    
        for (int i = 1; i <= SUM; i++)
        {
            //创建一个BMP图片
            sprintf(name, "C:\\Users\\Administrator\\Desktop\\data\\image\\%05d.bmp", i);
            FILE *ftmp = fopen(name, "wb");
            if (ftmp == NULL)
                printf("Create %05d.bmp failure\n", i);
            else
            {
                //读取一张图片的位图信息
                fread(image, sizeof(char), 28 * 28, fp1);
                //写文件头
                fwrite(&fh, sizeof(fileHeader), 1, ftmp);
                //写信息头
                fwrite(&fi, sizeof(fileInfo), 1, ftmp);
                //写调色板
                fwrite(fq, sizeof(rgbq), 256, ftmp);
                //写位图信息
                fwrite(image, fi.biSizeImage, 1, ftmp);
                fclose(ftmp);
            }
        }
        fclose(fp1);
        printf("------------------------------\n");
    
        return 0;
    }

**结果如下：**  
![这里写图片描述][SouthEast]

[http_yann.lecun.com_exdb_mnist]: http://yann.lecun.com/exdb/mnist/
[SouthEast]: /images/20220612/5698fe7ae3e74e99af29a38f4fcdf3d1.png