Skip to content
main
Go to file
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
SCG
 
 
mPb
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Matlab implementation of Contour Integration using Graph-Cut and Non-Classical Receptive Field

Overview

We proposed a graph-based framework that gets the soft-value of other methods as its input and creates more meaningful contours. Inspired by the concept of non-classical receptive fields in the primary visual cortex, we considered important factors such as connectivity, smoothness, and length of the contour beside the soft-values.

alt tag

How to use

Test

  • Download the maxflow folder from this repository and make it. (The Original reference is: https://www.mathworks.com/matlabcentral/fileexchange/21310-maxflow?s_tid=mwa_osa_a)
  • The function edge2contour gets the soft edge-map and the parameters and returns the binary contour-map.
  • Run the Gradient_Magnitude_test.m, mPb_test.m and SCG_test.m to see the result of applying our framework on three methods including: Gradient Magnitude, mPb and SCG. The parameters are trained for these three methods. If you require to employ our framework on your soft method, you should train the parameters anew.

Train

  • Download the MCS library and its dependencies from https://www.mat.univie.ac.at/~neum/software/mcs/.
  • Download Berkeley Contour Detection and Image Segmentation Resources from https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_full.tgz and build it. It contains the BSDS500 dataset and also evaluation codes.
  • Run your soft edge detection method on training data and save the outputs as .mat format in a directory named softmap_directory.
  • Consider an initial threshold and convert soft outputs to binary. After that compute the direction map for each binary image and save them as .mat format in a directory named dirmap_directory. For this purpose, you can use skeletonOrientation.m function that gets a binary image and returns orientation. Keep in mind that the initial threshold should not be too high.
  • Open train.m file and edit the following paths: softmap_directory, dirmap_directory, training images directory, and ground truth images directory.
  • Open train.m file and set the path of required libraries and run it. You can easily change the optimization parameters.

About

Contour Integration using Graph-Cut and Non-Classical Receptive Field

Resources

Releases

No releases published

Packages

No packages published
You can’t perform that action at this time.