Structured forests for fast edge detection

Introduction

In this tutorial you will learn how to use structured forests for the purpose of edge detection in an image.

Examples

image

image

image

image

image

image

image

image

image

image

image

image

Note

binarization techniques like Canny edge detector are applicable to edges produced by both algorithms (Sobel and StructuredEdgeDetection::detectEdges).

Source Code

/**************************************************************************************
The structured forests for fast edge detection demo requires you to provide a model.
This model can be found at the opencv_extra repository on Github on the following link:
https://github.com/opencv/opencv_extra/blob/master/testdata/cv/ximgproc/model.yml.gz
***************************************************************************************/

#include <opencv2/ximgproc.hpp>
#include "opencv2/highgui.hpp"
#include <iostream>

using namespace cv;
using namespace cv::ximgproc;

const char* keys =
{
 "{i || input image file name}"
 "{m || model file name}"
 "{o || output image file name}"
};

int main( int argc, const char** argv )
{
 CommandLineParser parser(argc, argv, keys);
 parser.about("This sample demonstrates usage of structured forests for fast edge detection");
 parser.printMessage();

 if ( !parser.check() )
 {
 parser.printErrors();
 return -1;
 }

 String modelFilename = parser.get<String>("m");
 String inFilename = parser.get<String>("i");
 String outFilename = parser.get<String>("o");

 Mat image = imread(inFilename, IMREAD_COLOR);
 if ( image.empty() )
 CV_Error(Error::StsError, String("Cannot read image file: ") + inFilename);

 if ( modelFilename.size() == 0)
 CV_Error(Error::StsError, String("Empty model name"));

 image.convertTo(image, DataType<float>::type, 1/255.0);

 TickMeter tm;
 tm.start();
 Ptr<StructuredEdgeDetection> pDollar =
 createStructuredEdgeDetection(modelFilename);

 tm.stop();
 std::cout << "createStructuredEdgeDetection() time : " << tm << std::endl;

 tm.reset();
 tm.start();
 Mat edges;
 pDollar->detectEdges(image, edges);
 tm.stop();
 std::cout << "detectEdges() time : " << tm << std::endl;

 tm.reset();
 tm.start();
 // computes orientation from edge map
 Mat orientation_map;
 pDollar->computeOrientation(edges, orientation_map);

 // suppress edges
 Mat edge_nms;
 pDollar->edgesNms(edges, orientation_map, edge_nms, 2, 0, 1, true);

 tm.stop();
 std::cout << "nms time : " << tm << std::endl;

 if ( outFilename.size() == 0 )
 {
 imshow("edges", edges);
 imshow("edges nms", edge_nms);
 waitKey(0);
 }
 else
 imwrite(outFilename, 255*edges);

 return 0;
}

Explanation

1. Load source color image

    Mat image = imread(inFilename, IMREAD_COLOR);
    if ( image.empty() )
    CV_Error(Error::StsError, String("Cannot read image file: ") + inFilename);

2. Convert source image to float [0;1] range

    image.convertTo(image, DataType<float>::type, 1/255.0);

3. Run main algorithm

    Ptr<StructuredEdgeDetection> pDollar =
    createStructuredEdgeDetection(modelFilename);

    Mat edges;
    pDollar->detectEdges(image, edges);

    // computes orientation from edge map
    Mat orientation_map;
    pDollar->computeOrientation(edges, orientation_map);
   
    // suppress edges
    Mat edge_nms;
    pDollar->edgesNms(edges, orientation_map, edge_nms, 2, 0, 1, true);

4. Show results

    if ( outFilename.size() == 0 )
    {
    imshow("edges", edges);
    imshow("edges nms", edge_nms);
    waitKey(0);
    }
    else
    imwrite(outFilename, 255*edges);

   Literature

For more information, refer to the following papers : [56] [135]