How opencv implements image skew correction

opencvHow to implement image skew correction

今天小编给大家分享一下opencvHow to realize the relevant knowledge points of image tilt correction,内容详细,逻辑清晰,相信大部分人都还太了解这方面的知识,所以分享这篇文章给大家参考一下,希望大家阅读完这篇文章后有所收获,下面我们一起来了解一下吧.

Today's task is to picture these two situations,My current situation is the same as pictured,The left image is slightly tilted to the left,Then I need to push the left side down for tilt correction.右图,Right tilt I need to push the right side down to a horizontal position for tilt correction.My scene doesn't have a large angle of tilt,It's probably not more than that45°的小角度.
Standard traditional image processing methods.The first is grayscale,二值化,取轮廓,Filter to get the target（面积最大的那个）,得到倾斜角度,转正.
The specific process is not described in detail,You can see it all in the code.效果图如下：

#include <iostream>#include"opencv2/opencv.hpp"using namespace std;using namespace cv;bool cmp_x(cv::Point pt1, cv::Point pt2){    return (pt1.x < pt2.x);}cv::Point2f get_mid_pt(cv::Point2f pt1, cv::Point2f pt2){    return cv::Point2f((pt1.x + pt2.x)/2.0,(pt1.y + pt2.y) / 2.0);}double get_point_angle(cv::Point2f pointO,cv::Point2f pointA){    double angle = 0;    cv::Point2f point;    double temp;    point = cv::Point2f((pointA.x - pointO.x), (pointA.y - pointO.y));    if ((0==point.x) && (0==point.y))    {        return 0;    }    if (0==point.x)    {        angle = 90;        return angle;    }    if (0==point.y)    {        angle = 0;        return angle;    }    temp = fabsf(float(point.y)/float(point.x));    temp = atan(temp);    temp = temp*180/CV_PI ;    if ((0<point.x)&&(0<point.y))    {        angle = 360 - temp;        return angle;    }    if ((0>point.x)&&(0<point.y))    {        angle = 360 - (180 - temp);        return angle;    }    if ((0<point.x)&&(0>point.y))    {        angle = temp;        return angle;    }    if ((0>point.x)&&(0>point.y))    {        angle = 180 - temp;        return angle;    }    printf("sceneDrawing :: getAngle error!");    return -1;}int RotateImage(const cv::Mat &src, double angle, cv::Mat &dst, cv::Mat &rot_matrix, bool crop = true, int flags = cv::INTER_NEAREST,                int borderMode = cv::BORDER_CONSTANT, const cv::Scalar &borderValue = cv::Scalar()){    if(0 == src.cols * src.rows) { return 0;}    cv::Point2f center(src.cols / 2.0f, src.rows / 2.0f);    rot_matrix = cv::getRotationMatrix2D(center, angle, 1.0);    if (crop) {        if (dst.data == NULL) {            dst = cv::Mat(src.rows, src.cols, src.type());        }    } else {        cv::Rect bbox = cv::RotatedRect(center, cv::Size2f(src.cols, src.rows), angle).boundingRect();        double *p = (double *) rot_matrix.data;        p[2] += bbox.width / 2.0 - center.x;        p[5] += bbox.height / 2.0 - center.y;        if (dst.rows != bbox.height || dst.cols != bbox.width) {            dst = cv::Mat(bbox.height, bbox.width, src.type());        }    }    cv::warpAffine(src, dst, rot_matrix, dst.size(), flags, borderMode, borderValue);    return 0;}int main(int argc, char *argv[]){    cv::Mat img = cv::imread("/data_1/everyday/0325/13.jpeg");    cv::Mat m_gray,m_bi;    cv::cvtColor(img,m_gray,CV_BGR2GRAY);    cv::threshold(m_gray,m_bi,100,255,THRESH_BINARY_INV);    vector<vector<Point>> contours;    vector<Vec4i> hierarchy;    findContours(m_bi,contours,hierarchy,RETR_TREE,CHAIN_APPROX_SIMPLE,Point());    RotatedRect rt_rot_max,rt_tmp;    int max_szie = -1;    for(int i=0;i<contours.size();i++)    {        rt_tmp = minAreaRect(Mat(contours[i]));        if(rt_tmp.size.area() > max_szie)        {            max_szie = rt_tmp.size.area();            rt_rot_max = rt_tmp;        }    }    std::vector<cv::Point2f> v_pt(4);    rt_rot_max.points(v_pt.data());    std::sort(v_pt.begin(),v_pt.end(),cmp_x);    cv::Point2f pt_left = get_mid_pt(v_pt[0], v_pt[1]);    cv::Point2f pt_right = get_mid_pt(v_pt[2], v_pt[3]);    double ang = get_point_angle(pt_left,pt_right);    std::cout<<"ang="<<ang<<std::endl;    cv::circle(img,v_pt[0],6,Scalar(50,12,189),3);    cv::circle(img,v_pt[1],6,Scalar(10,255,255),3);    cv::circle(img,v_pt[2],6,Scalar(150,120,19),3);    cv::circle(img,v_pt[3],6,Scalar(0,0,0),3);    cv::Mat m_rot,rot_matrix;    RotateImage(img, -ang, m_rot, rot_matrix, false);    cv::imshow("m_rot",m_rot);    cv::imshow("m_bi",m_bi);    cv::imshow("m_gray",m_gray);    cv::imshow("img",img);    cv::waitKey(0);    return 0;}

以上就是“opencvHow to implement image skew correction”这篇文章的所有内容,感谢各位的阅读！相信大家阅读完这篇文章都有很大的收获,小编每天都会为大家更新不同的知识,如果还想学习更多的知识,请关注亿速云行业资讯频道.