opencv3/C++ 机器学习-最邻近算法KNN识别字符

如图，有如下字母表；

现尝试采用最邻近算法KNN，取前10列字符作为训练数据，然后识别字母表中的字符。

创建训练数据

首先通过获取前10列字符的轮廓外接矩形，将字符裁剪出作为训练样本建立图库。

#include <string>
#include <iostream>
#include <opencv2\opencv.hpp>
#include<numeric>
using namespace cv;


int main()
{
    Mat src, dst;
    src = imread("E:/image/image/letters.png");
    if (src.empty())
    {
        std::cout<<"can not load image \n"<<std::endl;
        return -1;
    }
    imshow("input", src);
    dst = src.clone();
    cvtColor(src, src, COLOR_BGR2GRAY); 
    Mat bin = src.clone();
    //创建感兴趣区域,选取左侧10列作为训练数据
    Mat ROI = src(Rect(0,0,400,src.size().height));
    blur(ROI, ROI, Size(9,9));
    threshold(ROI, ROI, 210, 255, THRESH_BINARY);
    Canny(ROI, ROI, 20, 80, 3, false);
    std::vector<std::vector<Point>> contours;
    std::vector<Vec4i>hierarchy;
    findContours(ROI, contours, hierarchy, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE, Point(0,0));
    RNG rng(0);
    std::vector<RotatedRect> minRects(contours.size());
    std::vector<float> height, width;
    for (int i = 0; i < contours.size(); i++)
    {
        Scalar color = Scalar(rng.uniform(0,255), rng.uniform(0,255), rng.uniform(0,255));
        Rect rect = boundingRect(contours[i]);
        height.push_back(rect.height);
        width.push_back(rect.width);
    }
    //获取字符外接矩形宽高的最大值
    std::vector<float>::iterator h = std::max_element(std::begin(height), std::end(height));
    std::vector<float>::iterator w = std::max_element(std::begin(width), std::end(width));
    //获得轮廓的外接矩形
    std::vector<Rect>rects;
    //将轮廓外接矩形按纵坐标排序
    std::vector<float>sequence;
    for (int i = 0; i < contours.size(); i++)
    {
        Scalar color = Scalar(0, rng.uniform(0,255), rng.uniform(0,255));
        Rect rect = boundingRect(contours[i]);
        Rect reRect = Rect(Point(rect.x + rect.width/2.0 - *w/2.0,rect.y + rect.height/2.0 - *h/2.0 + 2), Point(rect.x + rect.width/2.0 + *w/2.0, rect.y + rect.height/2.0 + *h/2.0 - 2));
        rectangle(dst, reRect, color, 2);   
        rects.push_back(reRect);
        sequence.push_back(reRect.y);
    }
    //按纵坐标排序
    sort(sequence.begin(), sequence.end());

    threshold(bin, bin, 0, 255, CV_THRESH_BINARY|CV_THRESH_OTSU);
    int n = 0;
    for (int i = 0; i < rects.size(); i++)
    {
        for (int j = 0; j < 26; j++)
        {
            std::string outPath = "E:/image/letters/";
            char label = 'a';
            char temp[256];
            if ((rects[i].y > sequence[j*10]-*h/2.0) && (rects[i].y < sequence[j*10]+*h/2.0))
            {
                label = label + j;
                sprintf_s(temp, "%d", n);
                outPath = outPath + label + "/" + temp + ".jpg";
                //std::cout<<outPath<<std::endl;
                imwrite(outPath,bin(rects[i]));
                n++;
            }
        }
    }

    imshow("output", dst);
    waitKey();

    return 0;
}

被选中的字符：

裁剪后得到的图库：

利用KNN识别字符

利用KNN识别出图中的字符，并将识别结果显示在图中对应的字符旁边。（同种字符用同一种颜色显示）

#include <io.h>
#include <string>
#include <iostream>
#include <opencv2\opencv.hpp>
#include <opencv2\ml.hpp>
#include<numeric>
using namespace cv;
using namespace ml;

int main()
{
    ===============================读取训练数据===============================
    //图片共有10类
    const int classSum = 26;
    //每类共50张图片
    const int imagesSum = 10;
    //图片尺寸
    const int imageRows = 33;
    const int imageCols = 26;
    //每一行一个训练图片
    float trainingData[classSum*imagesSum][imageRows*imageCols] = {{0}};
    //训练样本标签
    float labels[classSum*imagesSum]={0};

    for (int i = 0; i < classSum; i++)
    {
        //目标文件夹路径
        std::string inPath = "E:\\image\\letters\\";
        char label = 'a';
        int k = 0;
        label = label + i;
        inPath = inPath + label + "\\*.jpg";
        //用于查找的句柄
        long handle;
        struct _finddata_t fileinfo;
        //第一次查找
        handle = _findfirst(inPath.c_str(),&fileinfo);
        if(handle == -1)
            return -1;
        do
        {
            //找到的文件的文件名
            std::string imgname = "E:/image/letters/";
            imgname = imgname + label + "/" + fileinfo.name;
            //std::cout<<imgname<<std::endl;
            Mat src = imread(imgname, 0);
            if (src.empty())
            {
                std::cout<<"can not load image \n"<<std::endl;
                return -1;
            }
            //序列化后放入作为样本矩阵的一行
            for(int j = 0; j<imageRows*imageCols; j++)
            {
                trainingData[i*imagesSum + k][j] = (float)src.data[j];
            }
            // 设置样本标签
            labels[i*imagesSum + k] = label;
            k++;

        } while (!_findnext(handle, &fileinfo));
        _findclose(handle);
    }
    //训练样本数据及对应标签
    Mat trainingDataMat(classSum*imagesSum, imageRows*imageCols, CV_32FC1, trainingData);
    Mat labelsMat(classSum*imagesSum, 1, CV_32FC1, labels);
    //std::cout<<trainingDataMat<<std::endl;
    //std::cout<<labelsMat<<std::endl;

    ===============================创建KNN模型===============================
    Ptr<KNearest>model = KNearest::create();
    model->setDefaultK(classSum);
    model->setIsClassifier(true);
    Ptr<TrainData>trainData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    model->train(trainData);

    //model->save("E:/image/KNearestModel.xml"); 
    ===============================预测部分===============================

    Mat src, dst;
    src = imread("E:/image/image/letters.png");
    if (src.empty())
    {
        std::cout<<"can not load image \n"<<std::endl;
        return -1;
    }
    dst = src.clone();
    //创建感兴趣区域,选取右侧10列作为预测数据
    cvtColor(src, src, COLOR_BGR2GRAY);

    blur(src, src, Size(9,9));
    threshold(src, src, 210, 255, THRESH_BINARY);
    Canny(src, src, 20, 80, 3, false);
    std::vector<std::vector<Point>> contours;
    std::vector<Vec4i>hierarchy;
    findContours(src, contours, hierarchy, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE, Point(0,0));

    for (int i = 0; i < contours.size(); i++)
    {
        Rect rect = boundingRect(contours[i]);
        //以矩形中心及指定的宽高作为字符区域
        Rect reRect = Rect(Point(rect.x + rect.width/2.0 - imageCols/2.0,rect.y + rect.height/2.0 - imageRows/2.0), Point(rect.x + rect.width/2.0 + imageCols/2.0, rect.y + rect.height/2.0 + imageRows/2.0));
        Mat sampleImg;
        cvtColor(dst, sampleImg, COLOR_BGR2GRAY);
        threshold(sampleImg, sampleImg, 0, 255, CV_THRESH_BINARY|CV_THRESH_OTSU);
        Mat sample = Mat::zeros(Size(imageCols,imageRows), sampleImg.type());

        float sampleData[imageRows*imageCols];
        int nub = 0;
        for (int r = 0; r < imageRows; r++)
        {
            for (int c = 0; c < imageCols; c++)
            {
                sampleData[nub] = sampleImg.at<uchar>(reRect.y+r,reRect.x+c);
                nub++;
            }
        }

        Mat sampleDataMat(1, imageRows*imageCols, CV_32FC1, sampleData);
        char f;
        f = model->predict(sampleDataMat);
        char temp[256];
        sprintf_s(temp, "%c", f);
        std::cout<<temp<<"\n"<<std::endl;
        std::string text(temp);
        RNG rng(f);
        Scalar color = Scalar(rng.uniform(0,255), rng.uniform(0,255), rng.uniform(0,255));
        putText(dst, text, Point(reRect.x,reRect.y + reRect.height),1,1.5,color,2);
    }
    imshow("output",dst);
    waitKey();
    return 0;
}

显示识别结果：

---

相关链接：

图片及代码下载