Remapping

Prev Tutorial: Object detection with Generalized Ballard and Guil Hough Transform

Next Tutorial: Affine Transformations

Goal

In this tutorial you will learn how to:

a. Use the OpenCV function cv::remap to implement simple remapping routines.

Theory

What is remapping?

• It is the process of taking pixels from one place in the image and locating them in another position in a new image.
• To accomplish the mapping process, it might be necessary to do some interpolation for non-integer pixel locations, since there will not always be a one-to-one-pixel correspondence between source and destination images.

• We can express the remap for every pixel location \((x,y)\) as:

  \[g(x,y) = f ( h(x,y) )\]

  where \(g()\) is the remapped image, \(f()\) the source image and \(h(x,y)\) is the mapping function that operates on \((x,y)\).

• Let's think in a quick example. Imagine that we have an image \(I\) and, say, we want to do a remap such that:

  \[h(x,y) = (I.cols - x, y )\]

  What would happen? It is easily seen that the image would flip in the \(x\) direction. For instance, consider the input image:

  

  observe how the red circle changes positions with respect to x (considering \(x\) the horizontal direction):

  
• In OpenCV, the function cv::remap offers a simple remapping implementation.

Code

• What does this program do?
  • Loads an image
  • Each second, apply 1 of 4 different remapping processes to the image and display them indefinitely in a window.
  • Wait for the user to exit the program

C++

• The tutorial code's is shown lines below. You can also download it from here

  
  #include "opencv2/imgcodecs.hpp"
  #include "opencv2/highgui.hpp"
  #include "opencv2/imgproc.hpp"
  #include <iostream>
  
  using namespace cv;
  
  void update_map( int &ind, Mat &map_x, Mat &map_y );
  
  int main(int argc, const char** argv)
  {
      CommandLineParser parser(argc, argv, "{@image |chicky_512.png|input image name}");
      std::string filename = parser.get<std::string>(0);
      Mat src = imread( samples::findFile( filename ), IMREAD_COLOR );
      if (src.empty())
      {
          std::cout << "Cannot read image: " << filename << std::endl;
          return -1;
      }
  
      Mat dst(src.size(), src.type());
      Mat map_x(src.size(), CV_32FC1);
      Mat map_y(src.size(), CV_32FC1);
  
      const char* remap_window = "Remap demo";
      namedWindow( remap_window, WINDOW_AUTOSIZE );
  
      int ind = 0;
      for(;;)
      {
          update_map(ind, map_x, map_y);
          remap( src, dst, map_x, map_y, INTER_LINEAR, BORDER_CONSTANT, Scalar(0, 0, 0) );
  
          imshow( remap_window, dst );
  
          char c = (char)waitKey( 1000 );
          if( c == 27 )
          {
              break;
          }
      }
      return 0;
  }
  
  void update_map( int &ind, Mat &map_x, Mat &map_y )
  {
      for( int i = 0; i < map_x.rows; i++ )
      {
          for( int j = 0; j < map_x.cols; j++ )
          {
              switch( ind )
              {
              case 0:
                  if( j > map_x.cols*0.25 && j < map_x.cols*0.75 && i > map_x.rows*0.25 && i < map_x.rows*0.75 )
                  {
                      map_x.at<float>(i, j) = 2*( j - map_x.cols*0.25f ) + 0.5f;
                      map_y.at<float>(i, j) = 2*( i - map_x.rows*0.25f ) + 0.5f;
                  }
                  else
                  {
                      map_x.at<float>(i, j) = 0;
                      map_y.at<float>(i, j) = 0;
                  }
                  break;
              case 1:
                  map_x.at<float>(i, j) = (float)j;
                  map_y.at<float>(i, j) = (float)(map_x.rows - i);
                  break;
              case 2:
                  map_x.at<float>(i, j) = (float)(map_x.cols - j);
                  map_y.at<float>(i, j) = (float)i;
                  break;
              case 3:
                  map_x.at<float>(i, j) = (float)(map_x.cols - j);
                  map_y.at<float>(i, j) = (float)(map_x.rows - i);
                  break;
              default:
                  break;
              } // end of switch
          }
      }
      ind = (ind+1) % 4;
  }

Java

• The tutorial code's is shown lines below. You can also download it from here

  import org.opencv.core.Core;
  import org.opencv.core.CvType;
  import org.opencv.core.Mat;
  import org.opencv.highgui.HighGui;
  import org.opencv.imgcodecs.Imgcodecs;
  import org.opencv.imgproc.Imgproc;
  
  class Remap {
      private Mat mapX = new Mat();
      private Mat mapY = new Mat();
      private Mat dst = new Mat();
      private int ind = 0;
  
      private void updateMap() {
          float buffX[] = new float[(int) (mapX.total() * mapX.channels())];
          mapX.get(0, 0, buffX);
  
          float buffY[] = new float[(int) (mapY.total() * mapY.channels())];
          mapY.get(0, 0, buffY);
  
          for (int i = 0; i < mapX.rows(); i++) {
              for (int j = 0; j < mapX.cols(); j++) {
                  switch (ind) {
                  case 0:
                      if( j > mapX.cols()*0.25 && j < mapX.cols()*0.75 && i > mapX.rows()*0.25 && i < mapX.rows()*0.75 ) {
                          buffX[i*mapX.cols() + j] = 2*( j - mapX.cols()*0.25f ) + 0.5f;
                          buffY[i*mapY.cols() + j] = 2*( i - mapX.rows()*0.25f ) + 0.5f;
                      } else {
                          buffX[i*mapX.cols() + j] = 0;
                          buffY[i*mapY.cols() + j] = 0;
                      }
                      break;
                  case 1:
                      buffX[i*mapX.cols() + j] = j;
                      buffY[i*mapY.cols() + j] = mapY.rows() - i;
                      break;
                  case 2:
                      buffX[i*mapX.cols() + j] = mapY.cols() - j;
                      buffY[i*mapY.cols() + j] = i;
                      break;
                  case 3:
                      buffX[i*mapX.cols() + j] = mapY.cols() - j;
                      buffY[i*mapY.cols() + j] = mapY.rows() - i;
                      break;
                  default:
                      break;
                  }
              }
          }
          mapX.put(0, 0, buffX);
          mapY.put(0, 0, buffY);
          ind = (ind+1) % 4;
      }
  
      public void run(String[] args) {
          String filename = args.length > 0 ? args[0] : "../data/chicky_512.png";
          Mat src = Imgcodecs.imread(filename, Imgcodecs.IMREAD_COLOR);
          if (src.empty()) {
              System.err.println("Cannot read image: " + filename);
              System.exit(0);
          }
  
          mapX = new Mat(src.size(), CvType.CV_32F);
          mapY = new Mat(src.size(), CvType.CV_32F);
  
          final String winname = "Remap demo";
          HighGui.namedWindow(winname, HighGui.WINDOW_AUTOSIZE);
  
          for (;;) {
              updateMap();
              Imgproc.remap(src, dst, mapX, mapY, Imgproc.INTER_LINEAR);
              HighGui.imshow(winname, dst);
              if (HighGui.waitKey(1000) == 27) {
                  break;
              }
          }
          System.exit(0);
      }
  }
  
  public class RemapDemo {
      public static void main(String[] args) {
          // Load the native OpenCV library
          System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
  
          new Remap().run(args);
      }
  }

Python

• The tutorial code's is shown lines below. You can also download it from here

  from __future__ import print_function
  import cv2 as cv
  import numpy as np
  import argparse
  
  
  def update_map(ind, map_x, map_y):
      if ind == 0:
          for i in range(map_x.shape[0]):
              for j in range(map_x.shape[1]):
                  if j > map_x.shape[1]*0.25 and j < map_x.shape[1]*0.75 and i > map_x.shape[0]*0.25 and i < map_x.shape[0]*0.75:
                      map_x[i,j] = 2 * (j-map_x.shape[1]*0.25) + 0.5
                      map_y[i,j] = 2 * (i-map_y.shape[0]*0.25) + 0.5
                  else:
                      map_x[i,j] = 0
                      map_y[i,j] = 0
      elif ind == 1:
          for i in range(map_x.shape[0]):
              map_x[i,:] = [x for x in range(map_x.shape[1])]
          for j in range(map_y.shape[1]):
              map_y[:,j] = [map_y.shape[0]-y for y in range(map_y.shape[0])]
      elif ind == 2:
          for i in range(map_x.shape[0]):
              map_x[i,:] = [map_x.shape[1]-x for x in range(map_x.shape[1])]
          for j in range(map_y.shape[1]):
              map_y[:,j] = [y for y in range(map_y.shape[0])]
      elif ind == 3:
          for i in range(map_x.shape[0]):
              map_x[i,:] = [map_x.shape[1]-x for x in range(map_x.shape[1])]
          for j in range(map_y.shape[1]):
              map_y[:,j] = [map_y.shape[0]-y for y in range(map_y.shape[0])]
  
  
  parser = argparse.ArgumentParser(description='Code for Remapping tutorial.')
  parser.add_argument('--input', help='Path to input image.', default='chicky_512.png')
  args = parser.parse_args()
  
  
  src = cv.imread(cv.samples.findFile(args.input), cv.IMREAD_COLOR)
  if src is None:
      print('Could not open or find the image: ', args.input)
      exit(0)
  
  
  
  map_x = np.zeros((src.shape[0], src.shape[1]), dtype=np.float32)
  map_y = np.zeros((src.shape[0], src.shape[1]), dtype=np.float32)
  
  
  
  window_name = 'Remap demo'
  cv.namedWindow(window_name)
  
  
  
  ind = 0
  while True:
      update_map(ind, map_x, map_y)
      ind = (ind + 1) % 4
      dst = cv.remap(src, map_x, map_y, cv.INTER_LINEAR)
      cv.imshow(window_name, dst)
      c = cv.waitKey(1000)
      if c == 27:
          break
  

Explanation

• Load an image:

C++

    Mat src = imread( samples::findFile( filename ), IMREAD_COLOR );
    if (src.empty())
    {
        std::cout << "Cannot read image: " << filename << std::endl;
        return -1;
    }

Java

        Mat src = Imgcodecs.imread(filename, Imgcodecs.IMREAD_COLOR);
        if (src.empty()) {
            System.err.println("Cannot read image: " + filename);
            System.exit(0);
        }

Python

src = cv.imread(cv.samples.findFile(args.input), cv.IMREAD_COLOR)
if src is None:
    print('Could not open or find the image: ', args.input)
    exit(0)

• Create the destination image and the two mapping matrices (for x and y )

C++

    Mat dst(src.size(), src.type());
    Mat map_x(src.size(), CV_32FC1);
    Mat map_y(src.size(), CV_32FC1);

Java

        mapX = new Mat(src.size(), CvType.CV_32F);
        mapY = new Mat(src.size(), CvType.CV_32F);

Python

map_x = np.zeros((src.shape[0], src.shape[1]), dtype=np.float32)
map_y = np.zeros((src.shape[0], src.shape[1]), dtype=np.float32)

• Create a window to display results

C++

    const char* remap_window = "Remap demo";
    namedWindow( remap_window, WINDOW_AUTOSIZE );

Java

        final String winname = "Remap demo";
        HighGui.namedWindow(winname, HighGui.WINDOW_AUTOSIZE);

Python

window_name = 'Remap demo'
cv.namedWindow(window_name)

• Establish a loop. Each 1000 ms we update our mapping matrices (mat_x and mat_y) and apply them to our source image:

C++

    int ind = 0;
    for(;;)
    {
        update_map(ind, map_x, map_y);
        remap( src, dst, map_x, map_y, INTER_LINEAR, BORDER_CONSTANT, Scalar(0, 0, 0) );

        imshow( remap_window, dst );

        char c = (char)waitKey( 1000 );
        if( c == 27 )
        {
            break;
        }
    }

Java

        for (;;) {
            updateMap();
            Imgproc.remap(src, dst, mapX, mapY, Imgproc.INTER_LINEAR);
            HighGui.imshow(winname, dst);
            if (HighGui.waitKey(1000) == 27) {
                break;
            }
        }

Python

ind = 0
while True:
    update_map(ind, map_x, map_y)
    ind = (ind + 1) % 4
    dst = cv.remap(src, map_x, map_y, cv.INTER_LINEAR)
    cv.imshow(window_name, dst)
    c = cv.waitKey(1000)
    if c == 27:
        break

• The function that applies the remapping is cv::remap . We give the following arguments:

  • src: Source image
  • dst: Destination image of same size as src
  • map_x: The mapping function in the x direction. It is equivalent to the first component of \(h(i,j)\)
  • map_y: Same as above, but in y direction. Note that map_y and map_x are both of the same size as src
  • INTER_LINEAR: The type of interpolation to use for non-integer pixels. This is by default.
  • BORDER_CONSTANT: Default

  How do we update our mapping matrices mat_x and mat_y? Go on reading:

• Updating the mapping matrices: We are going to perform 4 different mappings:
  1. Reduce the picture to half its size and will display it in the middle:

     \[h(i,j) = ( 2 \times i - src.cols/2 + 0.5, 2 \times j - src.rows/2 + 0.5)\]

     for all pairs \((i,j)\) such that: \(\dfrac{src.cols}{4}<i<\dfrac{3 \cdot src.cols}{4}\) and \(\dfrac{src.rows}{4}<j<\dfrac{3 \cdot src.rows}{4}\)
  2. Turn the image upside down: \(h( i, j ) = (i, src.rows - j)\)
  3. Reflect the image from left to right: \(h(i,j) = ( src.cols - i, j )\)
  4. Combination of b and c: \(h(i,j) = ( src.cols - i, src.rows - j )\)

This is expressed in the following snippet. Here, map_x represents the first coordinate of h(i,j) and map_y the second coordinate.

C++

void update_map( int &ind, Mat &map_x, Mat &map_y )
{
    for( int i = 0; i < map_x.rows; i++ )
    {
        for( int j = 0; j < map_x.cols; j++ )
        {
            switch( ind )
            {
            case 0:
                if( j > map_x.cols*0.25 && j < map_x.cols*0.75 && i > map_x.rows*0.25 && i < map_x.rows*0.75 )
                {
                    map_x.at<float>(i, j) = 2*( j - map_x.cols*0.25f ) + 0.5f;
                    map_y.at<float>(i, j) = 2*( i - map_x.rows*0.25f ) + 0.5f;
                }
                else
                {
                    map_x.at<float>(i, j) = 0;
                    map_y.at<float>(i, j) = 0;
                }
                break;
            case 1:
                map_x.at<float>(i, j) = (float)j;
                map_y.at<float>(i, j) = (float)(map_x.rows - i);
                break;
            case 2:
                map_x.at<float>(i, j) = (float)(map_x.cols - j);
                map_y.at<float>(i, j) = (float)i;
                break;
            case 3:
                map_x.at<float>(i, j) = (float)(map_x.cols - j);
                map_y.at<float>(i, j) = (float)(map_x.rows - i);
                break;
            default:
                break;
            } // end of switch
        }
    }
    ind = (ind+1) % 4;
}

Java

    private void updateMap() {
        float buffX[] = new float[(int) (mapX.total() * mapX.channels())];
        mapX.get(0, 0, buffX);

        float buffY[] = new float[(int) (mapY.total() * mapY.channels())];
        mapY.get(0, 0, buffY);

        for (int i = 0; i < mapX.rows(); i++) {
            for (int j = 0; j < mapX.cols(); j++) {
                switch (ind) {
                case 0:
                    if( j > mapX.cols()*0.25 && j < mapX.cols()*0.75 && i > mapX.rows()*0.25 && i < mapX.rows()*0.75 ) {
                        buffX[i*mapX.cols() + j] = 2*( j - mapX.cols()*0.25f ) + 0.5f;
                        buffY[i*mapY.cols() + j] = 2*( i - mapX.rows()*0.25f ) + 0.5f;
                    } else {
                        buffX[i*mapX.cols() + j] = 0;
                        buffY[i*mapY.cols() + j] = 0;
                    }
                    break;
                case 1:
                    buffX[i*mapX.cols() + j] = j;
                    buffY[i*mapY.cols() + j] = mapY.rows() - i;
                    break;
                case 2:
                    buffX[i*mapX.cols() + j] = mapY.cols() - j;
                    buffY[i*mapY.cols() + j] = i;
                    break;
                case 3:
                    buffX[i*mapX.cols() + j] = mapY.cols() - j;
                    buffY[i*mapY.cols() + j] = mapY.rows() - i;
                    break;
                default:
                    break;
                }
            }
        }
        mapX.put(0, 0, buffX);
        mapY.put(0, 0, buffY);
        ind = (ind+1) % 4;
    }

Python

def update_map(ind, map_x, map_y):
    if ind == 0:
        for i in range(map_x.shape[0]):
            for j in range(map_x.shape[1]):
                if j > map_x.shape[1]*0.25 and j < map_x.shape[1]*0.75 and i > map_x.shape[0]*0.25 and i < map_x.shape[0]*0.75:
                    map_x[i,j] = 2 * (j-map_x.shape[1]*0.25) + 0.5
                    map_y[i,j] = 2 * (i-map_y.shape[0]*0.25) + 0.5
                else:
                    map_x[i,j] = 0
                    map_y[i,j] = 0
    elif ind == 1:
        for i in range(map_x.shape[0]):
            map_x[i,:] = [x for x in range(map_x.shape[1])]
        for j in range(map_y.shape[1]):
            map_y[:,j] = [map_y.shape[0]-y for y in range(map_y.shape[0])]
    elif ind == 2:
        for i in range(map_x.shape[0]):
            map_x[i,:] = [map_x.shape[1]-x for x in range(map_x.shape[1])]
        for j in range(map_y.shape[1]):
            map_y[:,j] = [y for y in range(map_y.shape[0])]
    elif ind == 3:
        for i in range(map_x.shape[0]):
            map_x[i,:] = [map_x.shape[1]-x for x in range(map_x.shape[1])]
        for j in range(map_y.shape[1]):
            map_y[:,j] = [map_y.shape[0]-y for y in range(map_y.shape[0])]

Result

1. After compiling the code above, you can execute it giving as argument an image path. For instance, by using the following image:

   

2. This is the result of reducing it to half the size and centering it:

   

3. Turning it upside down:

   

4. Reflecting it in the x direction:

   

5. Reflecting it in both directions: