Opencv 基本操作七 提取两幅图像的重叠区域

使用opencv提取连个图像的重叠区域,其本质就是提取两个图像的特征点,然后对两个图像的特征点进行匹配,根据匹配的特征点计算出透视变换矩阵H,然后根据H即可提取出两个图像的重叠区域。这里要注意的是,普通的opencv库没有包含opencv-contrib,无法使用xfeatures2d.hpp里面的SURF、SIFT算法提取图像的特征点。故此,需要自行编译opencv(将opencv-contrib加到动态库中),或者下载别人编译好的opencv。

1、包含opencv-contrib的opencv

win10下vs2019编译的opencv470:https://download.csdn.net/download/a486259/87355761
win10下vs2019编译的opencv453+cuda:
https://download.csdn.net/download/a486259/81328226

2、核心步骤

提取图像重叠区域的关键步骤为,加载图像、提特征点、特征点匹配、计算透视变换矩阵、计算重叠区域、提取重叠区域。

2.1 加载图像

    stringstream fmt1, fmt2;
    fmt1 << "IMG_20221231_160033.jpg";
    Mat left = imread(fmt1.str());//左侧:图片路径
    fmt2 << "IMG_20221231_160037.jpg";
    Mat right = imread(fmt2.str());//右侧:图片路径
    Size msize =  { 512, 512 };//left.size();//
    resize(left, left, msize);
    resize(right, right, msize);

2.2 提取图像特征点

创建surf对象,提取两个图像的特征点和特征描述符

    Ptr<SURF>surf;   //可以容纳800个特征点
    surf = SURF::create(800);//参数 查找的海森矩阵 create 海森矩阵阀值
    vector<KeyPoint>key1, key2;//特征点
    Mat c, d;//特征点描述符
    //提取特征点
    surf->detectAndCompute(left, Mat(), key2, d);
    surf->detectAndCompute(right, Mat(), key1, c);

2.2 提取图像特征点

使用暴力匹配器匹配两个图像中的特征描述符,匹配结果存入 vector<DMatch>中,然后按照比例提取一定匹配的点

    BFMatcher matcher; //暴力匹配器
    vector<DMatch>matches;//DMatch 点和点之间的关系
    //使用暴力匹配器匹配特征点,找到存来
    matcher.match(d, c, matches);

    //3、筛选特征点
    //排序 从小到大
    sort(matches.begin(), matches.end());
    //保留最优的特征点对象
    vector<DMatch>good_matches;//最优
    //设置比例
    int ptrPoint = std::max(50, (int)(matches.size() * 0.15));
    for (int i = 0; i < ptrPoint; i++)
    {
        good_matches.push_back(matches[i]);
    }

    //4、最佳匹配的特征点连成线
    Mat outimg;
    drawMatches(left, key2, right, key1, good_matches, outimg,
        Scalar::all(-1), Scalar::all(-1),
        vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);

2.3 计算透视变换矩阵

提取vector<DMatch>中已匹配的特征点存入imagepoint1, imagepoint2中计算透视变换矩阵

//5、计算透视变换矩阵
    //提取已匹配的特征点
    vector<Point2f>imagepoint1, imagepoint2;
    for (int i = 0; i < good_matches.size(); i++)
    {
        imagepoint1.push_back(key1[good_matches[i].trainIdx].pt);
        imagepoint2.push_back(key2[good_matches[i].queryIdx].pt);
    }
    //透视转换
    Mat H = findHomography(imagepoint1, imagepoint2, cv::RANSAC);
    cout << "H:::" << H << endl;

2.4 计算透视角点

根据H矩阵计算出透视角点,并将透视检点绘制为mask

    //定义四个角点坐标。
    Point2f obj_corners[4] = { cv::Point(0,0),cv::Point(left.cols, 0), cv::Point(left.cols, left.rows), cv::Point(0, left.rows) };
    Point scene_corners2[4];  //在srcImage1上画线
    //获取透视变化的角点
    for (int i = 0; i < 4; i++)
    {
        double x = obj_corners[i].x;
        double y = obj_corners[i].y;

        double Z = 1. / (H.at<double>(2, 0) * x + H.at<double>(2, 1) * y + H.at<double>(2, 2));
        double X = (H.at<double>(0, 0) * x + H.at<double>(0, 1) * y + H.at<double>(0, 2)) * Z;
        double Y = (H.at<double>(1, 0) * x + H.at<double>(1, 1) * y + H.at<double>(1, 2)) * Z;
        scene_corners2[i] = cv::Point(cvRound(X), cvRound(Y));
    }

    //使用drawContours将透视检点绘制为mask(超出图像大小的区域不可见)
    Mat roi = Mat::zeros(left.size(), CV_8UC1);
    vector<vector<Point>> contour;
    vector<Point> pts;
    pts.push_back(scene_corners2[0]);
    pts.push_back(scene_corners2[1]);
    pts.push_back(scene_corners2[2]);
    pts.push_back(scene_corners2[3]);
    contour.push_back(pts);
    drawContours(roi, contour, 0, Scalar::all(255), -1);

2.5 截取重叠区域

截取两个图像的重叠区域并绘图展示

//截取第一个图的重复区域
    Mat dstimg1;
    left.copyTo(dstimg1, roi);
    Mat roi_left_by = get_by(roi);
    //截取第二个图的重复区域
    Mat dstimg2 = Mat::zeros(left.size(), CV_8UC3);
    warpPerspective(right, dstimg2, H, left.size());
    //对roi进行逆变换,得到在右图的重叠区域
    Mat roi_right;
    warpPerspective(roi, roi_right, H, left.size(), WARP_INVERSE_MAP);
    Mat roi_right_by = get_by(roi_right);

    imshow("outimg", outimg);
    imshows("ss", {left,dstimg1,roi_left_by+ left ,right,dstimg2,roi_right_by+ right }, 3);
    waitKey();

3、完整示例

3.1 完整代码

其中Mattool.hpp的内容在3.2章

#include <iostream>
#include <opencv2/opencv.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/xfeatures2d.hpp>
#include <opencv2/calib3d.hpp>
#include <opencv2/imgproc.hpp>
#include "Matutils.hpp"

using namespace std;
using namespace cv;
using namespace cv::xfeatures2d;

Mat sobel_mat(Mat srcImage) {
    Mat grad_x, grad_y;
    Mat abs_grad_x, abs_grad_y;
    //    计算x方向的梯度
    Sobel(srcImage, grad_x, CV_16S, 1, 0, 3, 1, 0, BORDER_DEFAULT);
    convertScaleAbs(grad_x, abs_grad_x);

    //    计算y方向的梯度
    Sobel(srcImage, grad_y, CV_16S, 0, 1, 3, 1, 0, BORDER_DEFAULT);
    convertScaleAbs(grad_y, abs_grad_y);

    //    合并梯度
    Mat dstImage;
    addWeighted(abs_grad_x, 0.5, abs_grad_y, 0.5, 0, dstImage);
    return dstImage;
}
Mat line2red(Mat line) {
    vector<cv::Mat> mv;
    Mat red_line;
    Mat zeros = Mat::zeros(line.size(), CV_8UC1);
    //mv.push_back(zeros);
    //mv.push_back(zeros);
    mv.push_back(line);
    mv.push_back(line);
    mv.push_back(line);
    //将vector内的多个mat合并为一个多通道mat
    cv::merge(mv, red_line);
    return red_line;
}
Mat get_by(Mat roi) {
    Mat roi_left, roi_erode, roi_by;
    Mat element = getStructuringElement(MORPH_RECT, Size(7, 7));
    roi_left = pad_mat(roi);
    morphologyEx(roi_left, roi_erode, MORPH_ERODE, element);
    roi_by = roi_left - roi_erode;
    roi_by = rm_pad(roi_by);
    //cvtColor(roi_by, roi_by, color);
    return line2red(roi_by);
}
void test_img() {
    stringstream fmt1, fmt2;
    fmt1 << "IMG_20221231_160033.jpg";
    Mat left = imread(fmt1.str());//左侧:图片路径
    fmt2 << "IMG_20221231_160037.jpg";
    Mat right = imread(fmt2.str());//右侧:图片路径
    Size msize =  { 512, 512 };//left.size();//
    resize(left, left, msize);
    resize(right, right, msize);

    //1、创建SURF对象
    Ptr<SURF>surf;   //可以容纳800个特征点
    surf = SURF::create(800);//参数 查找的海森矩阵 create 海森矩阵阀值
    vector<KeyPoint>key1, key2;//特征点
    Mat c, d;//特征点描述符
    //提取特征点
    surf->detectAndCompute(left, Mat(), key2, d);
    surf->detectAndCompute(right, Mat(), key1, c);

    //2、进行特征匹配
    BFMatcher matcher; //暴力匹配器
    //特征点对比,保存下来
    vector<DMatch>matches;//DMatch 点和点之间的关系
    //使用暴力匹配器匹配特征点,找到存来
    matcher.match(d, c, matches);

    //3、筛选特征点
    //排序 从小到大
    sort(matches.begin(), matches.end());
    //保留最优的特征点对象
    vector<DMatch>good_matches;//最优
    //设置比例
    int ptrPoint = std::max(50, (int)(matches.size() * 0.15));
    for (int i = 0; i < ptrPoint; i++)
    {
        good_matches.push_back(matches[i]);
    }

    //4、最佳匹配的特征点连成线
    Mat outimg;
    drawMatches(left, key2, right, key1, good_matches, outimg,
        Scalar::all(-1), Scalar::all(-1),
        vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);


    //5、进行透视变化
    //特征点配对
    vector<Point2f>imagepoint1, imagepoint2;
    for (int i = 0; i < good_matches.size(); i++)
    {
        imagepoint1.push_back(key1[good_matches[i].trainIdx].pt);
        imagepoint2.push_back(key2[good_matches[i].queryIdx].pt);
    }
    //透视转换
    Mat H = findHomography(imagepoint1, imagepoint2, cv::RANSAC);
    cout << "H:::" << H << endl;

    //根据H矩阵计算出透视角点
    //定义四个角点坐标。
    Point2f obj_corners[4] = { cv::Point(0,0),cv::Point(left.cols, 0), cv::Point(left.cols, left.rows), cv::Point(0, left.rows) };
    Point scene_corners2[4];  //在srcImage1上画线
    //获取透视变化的角点
    for (int i = 0; i < 4; i++)
    {
        double x = obj_corners[i].x;
        double y = obj_corners[i].y;

        double Z = 1. / (H.at<double>(2, 0) * x + H.at<double>(2, 1) * y + H.at<double>(2, 2));
        double X = (H.at<double>(0, 0) * x + H.at<double>(0, 1) * y + H.at<double>(0, 2)) * Z;
        double Y = (H.at<double>(1, 0) * x + H.at<double>(1, 1) * y + H.at<double>(1, 2)) * Z;
        scene_corners2[i] = cv::Point(cvRound(X), cvRound(Y));
    }

    //使用drawContours将透视检点绘制为mask(超出图像大小的区域不可见)
    Mat roi = Mat::zeros(left.size(), CV_8UC1);
    vector<vector<Point>> contour;
    vector<Point> pts;
    pts.push_back(scene_corners2[0]);
    pts.push_back(scene_corners2[1]);
    pts.push_back(scene_corners2[2]);
    pts.push_back(scene_corners2[3]);
    contour.push_back(pts);
    drawContours(roi, contour, 0, Scalar::all(255), -1);

    //截取第一个图的重复区域
    Mat dstimg1;
    left.copyTo(dstimg1, roi);
    Mat roi_left_by = get_by(roi);


    //截取第二个图的重复区域
    Mat dstimg2 = Mat::zeros(left.size(), CV_8UC3);
    warpPerspective(right, dstimg2, H, left.size());
    //对roi进行逆变换,得到在右图的重叠区域
    Mat roi_right;
    warpPerspective(roi, roi_right, H, left.size(), WARP_INVERSE_MAP);
    Mat roi_right_by = get_by(roi_right);

    imshow("outimg", outimg);
    imshows("ss", {left,dstimg1,roi_left_by+ left ,right,dstimg2,roi_right_by+ right }, 3);
    waitKey();
}
int main(int argc, char* argv[])
{
    test_img();
    return 0;
}

3.2 Matutils.hpp

这是进行图像处理中通用的函数库

#ifndef __Matutils__
#define __Matutils__
#pragma once
#include <iostream>
#include <opencv2/opencv.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgproc.hpp>

#include <iostream>  
#include <vector>
#include <io.h>
#include <stdlib.h>
#include <iostream>
#include <string>
using namespace std;
using namespace cv;
//对图像填充黑边
inline  Mat pad_mat(Mat img,int pad=10) {
    Mat imgWindow = Mat::zeros(img.cols + 2*pad , img.rows + 2 * pad, img.type());
    img.copyTo(imgWindow(Rect({ pad ,pad }, img.size())));
    return imgWindow;
}
//移除图像填充的黑边
inline  Mat rm_pad(Mat img, int pad = 10) {
    Size size = { img.cols - 2 * pad , img.rows - 2 * pad };
    return img(Rect({ pad ,pad }, size));
}
//图像拼接
inline  void multipleImage(vector<Mat> imgVector, Mat& dst, int imgCols, int MAX_PIXEL = 300)
{
    //两列图像间的空白区域
    int pad = 10;
    int imgNum = imgVector.size();
    //选择图片最大的一边 将最大的边按比例变为300像素
    Size imgOriSize = imgVector[0].size();
    int imgMaxPixel = max(imgOriSize.height, imgOriSize.width);
    //获取最大像素变为MAX_PIXEL的比例因子
    double prop = imgMaxPixel < MAX_PIXEL ? (double)imgMaxPixel / MAX_PIXEL : MAX_PIXEL / (double)imgMaxPixel;
    Size imgStdSize(imgOriSize.width * prop, imgOriSize.height * prop); //窗口显示的标准图像的Size

    Mat imgStd; //标准图片
    Point2i location(0, 0); //坐标点(从0,0开始)
    //Mat imgWindow(imgStdSize.height * ((imgNum - 1) / imgCols + 1), imgStdSize.width * imgCols+ pad * imgCols-pad, imgVector[0].type());
    int imgRows = (imgNum - 1) / imgCols + 1;
    Mat imgWindow = Mat::zeros(imgStdSize.height * imgRows + pad * imgRows - pad, imgStdSize.width * imgCols + pad * imgCols - pad, imgVector[0].type());
    for (int i = 0; i < imgNum; i++)
    {
        location.x = (i % imgCols) * (imgStdSize.width + pad);
        location.y = (i / imgCols) * imgStdSize.height;
        resize(imgVector[i], imgStd, imgStdSize, prop, prop, INTER_LINEAR); //设置为标准大小

        //将imgStd复制到imgWindow的指定区域中
        imgStd.copyTo(imgWindow(Rect(location, imgStdSize)));
    }
    dst = imgWindow;
}
//多图显示
inline  void imshows(string title, vector<Mat> imgVector,  int imgCols = -1) {
    Mat dst;
    if (imgCols == -1) {
        imgCols = imgVector.size();
    }
    multipleImage(imgVector, dst, imgCols);
    namedWindow(title);
    imshow(title, dst);
    imwrite(title + ".png", dst);
}
//删除小面积的连通域
inline  Mat deleteMinWhiteArea(Mat src, int min_area) {
    Mat labels, stats, centroids, img_color, grayImg;
    //1、连通域信息统计
    int nccomps = connectedComponentsWithStats(
        src, //二值图像
        labels,
        stats,
        centroids
    );

    //2、连通域状态区分
    //为每一个连通域初始化颜色表
    vector<Vec3b> colors(nccomps);
    colors[0] = Vec3b(0, 0, 0); // background pixels remain black.
    for (int i = 1; i < nccomps; i++)
    {
        colors[i] = Vec3b(rand() % 256, rand() % 256, rand() % 256);
        //面积阈值筛选
        if ((stats.at<int>(i, CC_STAT_AREA) < min_area))
        {
            //如果连通域面积不合格则置黑
            colors[i] = Vec3b(0, 0, 0);
        }
    }
    //3、连通域删除
    //按照label值,对不同的连通域进行着色
    img_color = Mat::zeros(src.size(), CV_8UC3);
    for (int y = 0; y < img_color.rows; y++)
    {
        int* labels_p = labels.ptr<int>(y);//使用行指针,加速运算
        Vec3b* img_color_p = img_color.ptr<Vec3b>(y);//使用行指针,加速运算
        for (int x = 0; x < img_color.cols; x++)
        {
            int label = labels_p[x];//取出label值
            CV_Assert(0 <= label && label <= nccomps);
            img_color_p[x] = colors[label];//设置颜色
        }
    }
    //return img_color;
    //如果是需要二值结果则将img_color进行二值化
    cvtColor(img_color, grayImg, COLOR_BGR2GRAY);
    threshold(grayImg, grayImg, 1, 255, THRESH_BINARY);
    return grayImg;
}
//删除图形中小面积的黑色孔洞
inline  Mat deleteMinBlackArea(Mat src, int min_area) {
    Mat inv = 255 - src;//颜色取反
    Mat res = deleteMinWhiteArea(inv, min_area);
    return 255 - res;//颜色取反
}
//找图中topk个连通域
inline  bool mypairsort(pair<int, int> i, pair<int, int> j) { return (i.second > j.second); }
inline  Mat findTopKArea(Mat srcImage, int topk)
{
    Mat temp;
    Mat labels;
    srcImage.copyTo(temp);

    //1. 标记连通域
    int n_comps = connectedComponents(temp, labels, 4, CV_16U);
    vector<pair<int, int>> histogram_of_labels;
    for (int i = 0; i < n_comps; i++)//初始化labels的个数为0
    {
        histogram_of_labels.push_back({ i,0 });
    }

    int rows = labels.rows;
    int cols = labels.cols;
    for (int row = 0; row < rows; row++) //计算每个labels的个数--即连通域的面积
    {
        for (int col = 0; col < cols; col++)
        {
            histogram_of_labels.at(labels.at<unsigned short>(row, col)).second += 1;
        }
    }
    //histogram_of_labels.at(0).second = 0; //将背景的labels个数设置为0

    //2.对连通域进行排序
    std::sort(histogram_of_labels.begin(), histogram_of_labels.end(), mypairsort);
    //3. 取前k个连通域的labels id
    vector<int> select_labels;
    for (int i = 0; i < topk; i++)
    {
        if (histogram_of_labels[i].first == 0) {
            topk += 1;
            //如果碰到背景,则跳过,且topk+1
        }
        else {
            select_labels.push_back(histogram_of_labels[i].first);
        }
    }

    //3. 将label id在select_labels的连通域标记为255,并将其他连通域置0
    for (int row = 0; row < rows; row++)
    {
        for (int col = 0; col < cols; col++)
        {
            int now_label_id = labels.at<unsigned short>(row, col);
            if (std::count(select_labels.begin(), select_labels.end(), now_label_id)) {
                labels.at<unsigned short>(row, col) = 255;
            }
            else {
                labels.at<unsigned short>(row, col) = 0;
            }
        }
    }

    //4. 将图像更改为CV_8U格式
    labels.convertTo(labels, CV_8U);
    return labels;
}
//获取脊线
inline Mat get_ridge_line(Mat dst, int ksize = 3) {
    Mat skeleton, result, open_dst;
    Mat kernel = getStructuringElement(MORPH_CROSS, Size(ksize, ksize));
    skeleton = Mat::zeros(dst.rows, dst.cols, dst.type());
    while (true) {
        if (sum(dst)[0] == 0) {
            break;
        }
        morphologyEx(dst, dst, MORPH_ERODE, kernel);//消除毛刺,删除部分连通域
        morphologyEx(dst, open_dst, MORPH_OPEN, kernel);
        result = dst - open_dst;
        skeleton = skeleton + result;
    }
    return skeleton;
}
#endif

3.3 测试效果

两个图像的特征点匹配关系
在这里插入图片描述
算法提取出的重叠区域(中间列),第一例为两个原图,最后一列为重叠区域在原图中的效果。
在这里插入图片描述

文章出处登录后可见!

已经登录?立即刷新

共计人评分,平均

到目前为止还没有投票!成为第一位评论此文章。

(0)
社会演员多的头像社会演员多普通用户
上一篇 2023年3月1日 上午8:32
下一篇 2023年3月1日 上午8:35

相关推荐