《图像处理实例》 之 寻找图纸标注

 

 

---

要求：寻找图纸零件所有标注的字符，包括位置信息+图像信息

方法：

　　1.利用形态学+轮廓信息去查询

　　　　这里精度不是很高，计算难度也比较复杂，好处是思想简单。

　　2.利用模板匹配　　　　

　　　　这里是保证模板和实际相差不大，不然匹配精度就很差了。

 

---

 

#include <opencv2/opencv.hpp>
#include <iostream>
#include "math.h"
using namespace cv;
using namespace std;

typedef struct MyStruct
{
    Rect my_rec;
    Mat my_img;
}MyStruct;

int main(int argc, char*argv[])
{
    Mat inputImage = imread("1.png");
    Mat showImage = inputImage.clone();
    cvtColor(inputImage, inputImage, CV_BGR2GRAY);
    Mat morph, gray = inputImage.clone(), showGray;
    showGray.create(inputImage.size(), CV_8UC1);
    showGray.setTo(0);
    vector<vector<Point>> contours;
    vector<Vec4i> hierarchy;
    findContours(inputImage, contours, hierarchy, RETR_TREE, CHAIN_APPROX_SIMPLE, Point(-1, -1));
    Rect rec_adapt;
    for (size_t i = 0; i < contours.size(); i++)
    {
        int x = minAreaRect(contours[i]).boundingRect().x;
        int y = minAreaRect(contours[i]).boundingRect().y;
        int width = minAreaRect(contours[i]).boundingRect().width;
        int height = minAreaRect(contours[i]).boundingRect().height;
        Mat true_image;
        int true_pix_count;
        double true_pix_rate;
        if (x < 0 || y < 0) true_pix_rate = 1;
        else
        {
            true_image = gray(Rect(x, y, width, height));
            true_pix_count = countNonZero(true_image);
            true_pix_rate = static_cast<double>(true_pix_count) / static_cast<double>(minAreaRect(contours[i]).boundingRect().area());
        }             
        double angle = minAreaRect(contours[i]).angle;
        bool flag_angle = (angle == 9 ||  angle == 180 || angle == 0 ) ? true : false;//|| angle == 270
        if (minAreaRect(contours[i]).size.height >= 10 && minAreaRect(contours[i]).size.height <= 20 && minAreaRect(contours[i]).size.width >= 4 && minAreaRect(contours[i]).size.width <= 30  && flag_angle && true_pix_rate <= 0.8)//
        {
            drawContours(showGray, contours, static_cast<int>(i), Scalar(255, 255, 255), 1);
        }
    }
    Mat img1;
    Mat kernel_x = getStructuringElement(MORPH_RECT, Size(20,1));
    Mat kernel_y = getStructuringElement(MORPH_RECT, Size(1, 28));
    Mat kernel_x_l = getStructuringElement(MORPH_RECT, Size(20, 1));
    morphologyEx(showGray, showGray, MORPH_DILATE, kernel_x);
    morphologyEx(showGray, showGray, MORPH_DILATE, kernel_x);
    morphologyEx(showGray, img1, MORPH_OPEN, kernel_y);
    showGray = showGray - img1;
    morphologyEx(showGray, showGray, MORPH_CLOSE, kernel_x_l);
    findContours(showGray, contours, hierarchy, RETR_TREE, CHAIN_APPROX_SIMPLE, Point(-1, -1));
    vector<MyStruct> my_class;
    for (size_t i = 0; i < contours.size(); i++)
    {    
        if (boundingRect(contours[i]).width > 60)
        {
            int x = minAreaRect(contours[i]).boundingRect().x;
            int y = minAreaRect(contours[i]).boundingRect().y;
            int width = minAreaRect(contours[i]).boundingRect().width;
            int height = minAreaRect(contours[i]).boundingRect().height;
            MyStruct Struct_temp;
            Struct_temp.my_rec = boundingRect(contours[i]);
            Struct_temp.my_img = showImage(Rect(x, y, width, height)).clone();
            my_class.push_back(Struct_temp);
            rectangle(showImage, boundingRect(contours[i]), Scalar(0, 0, 255));
        }        
    }
    

    waitKey(0);
    return 0;

}

 

效果图不上了，不想再去运行了