类的格式按照bmp的文件头、信息头、调色板和数据域写好即可
因为图片是按字节读的，所以要强制让编译器不对结构体进行字节对齐，否则会出错
因为bmp格式规定存储时每行的像素数要补齐至4的整数倍，所以读入和写出时要加一些对应的操作
——————————————————————————————————————————————
2019.9.6更新
增加了边缘检测、均值滤波和中值滤波
详情见注释

#include <bits/stdc++.h>#pragma GCC optimize(3)
using namespace std;typedef unsigned short w;
typedef unsigned int dw;class bmp
{
#pragma pack(1)//取消编译器的字节对齐struct FileHead{w type;dw size;w reserved1, reserved2;dw offset;} fileHead{};#pragma pack(1)struct InfoHead{dw size, width, height;w planes; //图像的位面数w bitNum; //每个像素的位数dw compression; //压缩类型dw imgSize; //单位为字节dw xPelsPerMeter, yPelsPerMeter;dw colorNum; //使用的色彩数dw colorImportantNum; //重要的颜色数} infoHead{};#pragma pack(1)struct Palette{w b, g, r, reversed;} palette{};struct Sum;
#pragma pack(1)struct Pixel{unsigned char b, g, r;//24位位图按照BGR的顺序存放， 32位BGRA
//        friend ostream &operator<<(ostream &out, const Pixel a)
//        {
//            return out << '[' << (unsigned) a.b << ' ' << (unsigned) a.g << ' ' << (unsigned) a.r << ']';
//        }Pixel(unsigned char b = 0, unsigned char g = 0, unsigned char r = 0) : b(b), g(g), r(r){}Pixel &operator=(const Sum &a){b = (unsigned char) a.b;g = (unsigned char) a.g;r = (unsigned char) a.r;}bool operator<(const Pixel &a) const{if (b == a.b){if (g == a.g)return r < a.r;return g < a.g;}return b < a.b;}};struct Sum  //用来维护二维前缀和的结构体{           // 因为pixel的rgb都是1字节，求和会溢出，所以另开一个int b, g, r;Sum(int b = 0, int g = 0, int r = 0) : b(b), g(g), r(r){}template<typename T>//既可以+=Sum， 也可以+=PixelSum &operator+=(const T &a){b += a.b;g += a.g;r += a.r;return *this;}Sum operator+(const Sum &a){return {b + a.b, g + a.g, r + a.r};}Sum operator-(const Sum &a){return {b - a.b, g - a.g, r - a.r};}template<typename T>Sum &operator-=(const T &a){b -= a.b;g -= a.g;r -= a.r;return *this;}Sum &operator/=(const int k){b /= k;g /= k;r /= k;}};inline int pos(int i, int j){return i * realWidth + j;}ifstream in;ofstream out;int realWidth{};
public:bmp() = default;~bmp(){delete[] img;}unsigned int n{};Pixel *img = nullptr;bool open(const string &file){in.open(file, ios::in | ios::binary);if (!in)return 0 & puts("无法打开文件");return true;}void read(){if (!in){puts("未打开文件");return;}in.read((char *) &fileHead, sizeof(fileHead));in.read((char *) &infoHead, sizeof(infoHead));realWidth = ((infoHead.width - 1) / 4) * 4 + 4;n = infoHead.height * realWidth;img = new Pixel[infoHead.height * realWidth];in.read((char *) img, sizeof(Pixel) * n);in.close();}Pixel &operator[](int id){return img[id];}const InfoHead &getInfoHead() const{return infoHead;}const FileHead &getFileHead() const{return fileHead;}void write(const string &file){out.open(file, ios::out | ios::binary);out.write((char *) &fileHead, sizeof(fileHead));out.write((char *) &infoHead, sizeof(infoHead));out.write((char *) img, sizeof(Pixel) * n);out.close();}void toGrayImg(){unsigned char val;for (int i = 0; i < n; ++i){val = 0.587 * img[i].g + 0.299 * img[i].r + 0.114 * img[i].b;img[i] = Pixel(val, val, val);//rgb都设为一样就是灰度图了}}void meanFiltering(int k)//k是半径, 如k取2, 求均值的范围为3 * 3{--k;//均值滤波， 对每个大小为(2 * k + 1) * (2 * k + 1)的子矩阵求和再平均，把中心的值改为均值。Sum *dp = new Sum[realWidth * infoHead.height];//这里用了二维前缀和来加速
#define summ(x1, y1, x2, y2) (dp[pos(x2, y2)]+(x1 && y1 ? dp[pos(x1-1, y1-1)] : 0)-(x1 ? dp[pos(x1-1, y2)] : 0)-(y1 ? dp[pos(x2, y1-1)] : 0))for (int i = 0; i < infoHead.height; ++i)//预处理二维前缀和{for (int j = 0; j < infoHead.width; ++j){dp[pos(i, j)] += img[pos(i, j)];if (i)dp[pos(i, j)] += dp[pos(i - 1, j)];if (j)dp[pos(i, j)] += dp[pos(i, j - 1)];if (i && j)dp[pos(i, j)] -= dp[pos(i - 1, j - 1)];}}Sum val;for (int i = 0; i < infoHead.height; ++i){for (int j = 0; j < infoHead.width; ++j){if (i < k || j < k || i + k > infoHead.height - 1 || j + k > infoHead.width - 1) continue;val = summ(i - k, j - k, i + k, j + k);val /= (k * 2 + 1) * (k * 2 + 1);img[pos(i, j)] = Pixel((unsigned char) val.b, (unsigned char) val.g, (unsigned char) val.r);}}delete[] dp;}Pixel getMedian(int x, int y, int k){vector<Pixel> a;a.clear();for (int i = x - k; i <= x + k; ++i)for (int j = y - k; j <= y + k; ++j)a.push_back(img[pos(i, j)]);
//        nth_element(a.begin(), a.begin() + ((2 * k + 1) * (2 * k + 1)) / 2, a.end());sort(a.begin(), a.end());return a[((2 * k + 1) * (2 * k + 1)) / 2];}void solve()//增加噪声，用于检验中值滤波和均值滤波{for (int i = 0; i < infoHead.height; ++i){for (int j = 0; j < infoHead.width; ++j){if (rand() < 1000)img[pos(i, j)] = Pixel(255, 255, 255);}}}void medianFiltering(int k)//k是半径{//中值滤波，用子矩阵的中位数替换中心的值Pixel *IMG = new Pixel[infoHead.height * ((infoHead.width - 1) / 4 * 4 + 4)];--k;for (int i = k; i + k < infoHead.height; ++i){for (int j = k; j + k < infoHead.width; ++j){IMG[pos(i, j)] = getMedian(i, j, k);}}for (int i = k; i + k < infoHead.height; ++i){for (int j = k; j + k < infoHead.width; ++j){img[pos(i, j)] = IMG[pos(i, j)];}}delete[]IMG;}void toEdge()//边缘检测{//采用梯度下降法，对x方向和y方向分别求差分，每个像素处保存一个向量(x, y)， 向量的模小于阈值的像素涂黑，其余涂白int *difR = new int[realWidth * infoHead.height], *difC = new int[realWidth * infoHead.height];for (int i = 1; i < infoHead.height; ++i){for (int j = 0; j < infoHead.width; ++j){difR[pos(i, j)] = img[pos(i, j)].b - img[pos(i - 1, j)].b;}}for (int i = 0; i < infoHead.height; ++i){for (int j = 1; j < infoHead.width; ++j){difC[pos(i, j)] = img[pos(i, j)].b - img[pos(i, j - 1)].b;}}for (int i = 0; i < n; ++i){if (sqrt(difR[i] * difR[i] + difC[i] * difC[i]) < 6)img[i] = Pixel(255, 255, 255);elseimg[i] = Pixel(0, 0, 0);}delete[] difC;delete[] difR;}
};int main()
{bmp pic;pic.open("test.bmp");pic.read();
//    printf("%u %u\n", pic.getInfoHead().imgSize, pic.getInfoHead().width);
//    pic.toGrayImg();
//    pic.solve();
//    pic.write("g01.bmp");pic.toEdge();//    pic.meanFiltering(2);pic.medianFiltering(2);pic.write("test.bmp");return 0;
}

原图

直接边缘检测的效果

中值滤波后的效果