在opencv c ++中旋转图像答案

【问题标题】：Rotate back an image in opencv c++在opencv c ++中旋转图像
【发布时间】：2017-05-17 11:34:22
【问题描述】：

我使用此代码在 OpenCV 中旋转我的图像：

// get rotation matrix for rotating the image around its center
Point2f center22(RGBsrc.cols/2.0, RGBsrc.rows/2.0);
Mat rot = getRotationMatrix2D(center22, Angle, 1.0);
// determine bounding rectangle
Rect bbox = RotatedRect(center22,RGBsrc.size(),Angle).boundingRect();
// adjust transformation matrix
rot.at<double>(0,2) += bbox.width/2.0 - center22.x;
rot.at<double>(1,2) += bbox.height/2.0 - center22.y;
Mat dst;
warpAffine(RGBsrc, dst, rot, bbox.size());
imshow("rotated_im", dst);

它工作正常。现在我想将该图像旋转回原始图像。当我使用下面的代码时，我看到图像中对象的位置与原始图像中的位置不同。为什么会这样？如何将图像旋转回来？

Point2f center22(RGBsrc.cols/2.0, RGBsrc.rows/2.0);
Mat rot2 = getRotationMatrix2D(center22, -Angle, 1.0);
Rect bbox2 = RotatedRect(center22,RGBsrc.size(), -Angle).boundingRect();
rot2.at<double>(0,2) += bbox2.width/2.0 - center22.x;
rot2.at<double>(1,2) += bbox2.height/2.0 - center22.y;
Mat Rotatedback;
warpAffine(RotatedRGB, Rotatedback, rot2, bbox2.size());

【问题讨论】：

标签： c++ image opencv image-rotation

【解决方案1】：

与其重新计算变换矩阵，不如简单地取它的逆矩阵？

// Invert the affine transformation
Mat rotInv;
cv::invertAffineTransform(rot, rotInv);

加载图片后：

您可以像现在一样旋转它：

// get rotation matrix for rotating the image around its center
Point2f center22(RGBsrc.cols / 2.0, RGBsrc.rows / 2.0);
Mat rot = getRotationMatrix2D(center22, Angle, 1.0);

// determine bounding rectangle
Rect bbox = RotatedRect(center22, RGBsrc.size(), Angle).boundingRect();

// adjust transformation matrix
rot.at<double>(0, 2) += bbox.width / 2.0 - center22.x;
rot.at<double>(1, 2) += bbox.height / 2.0 - center22.y;
Mat RotatedRGB;
warpAffine(RGBsrc, RotatedRGB, rot, bbox.size());

然后计算逆变换矩阵并回退：

// Invert the affine transformation
Mat rotInv;
cv::invertAffineTransform(rot, rotInv);

// Get back the original image
Mat Rotatedback;
warpAffine(RotatedRGB, Rotatedback, rotInv, Size(RGBsrc.cols, RGBsrc.rows));

完整代码供参考：

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

int main()
{
    Mat3b RGBsrc = imread("path_to_image");
    double Angle = 30.0;

    // get rotation matrix for rotating the image around its center
    Point2f center22(RGBsrc.cols / 2.0, RGBsrc.rows / 2.0);
    Mat rot = getRotationMatrix2D(center22, Angle, 1.0);

    // determine bounding rectangle
    Rect bbox = RotatedRect(center22, RGBsrc.size(), Angle).boundingRect();

    // adjust transformation matrix
    rot.at<double>(0, 2) += bbox.width / 2.0 - center22.x;
    rot.at<double>(1, 2) += bbox.height / 2.0 - center22.y;
    Mat RotatedRGB;
    warpAffine(RGBsrc, RotatedRGB, rot, bbox.size());

    // Invert the affine transformation
    Mat rotInv;
    cv::invertAffineTransform(rot, rotInv);

    // Get back the original image
    Mat Rotatedback;
    warpAffine(RotatedRGB, Rotatedback, rotInv, Size(RGBsrc.cols, RGBsrc.rows));

    imshow("Original", RGBsrc);
    imshow("Rotated", RotatedRGB);
    imshow("Rotated Back", Rotatedback);
    waitKey();

    return 0;
}

【讨论】：

【解决方案2】：

如果您在调用warpAffine() 时添加WARP_INVERSE_MAP 标志，它将应用逆变换矩阵，让您回到原始图像。例如，您应该能够将问题中的最后一行代码替换为以下内容：

warpAffine(RotatedRGB, Rotatedback, rot, RGBsrc.size(), INTER_LINEAR|WARP_INVERSE_MAP);

注意：我还没有测试过这段代码，但是我用WARP_INVERSE_MAP 对warpPerspective() 做了同样的事情，而且效果很好。

文档： https://docs.opencv.org/3.4/da/d54/group__imgproc__transform.html#ga0203d9ee5fcd28d40dbc4a1ea4451983

【讨论】：