This code uses opencv 3 libraries
CMake and basic compilation tools are needed for this to work.
git clone https://github.com/jimmylomro/own.git
cd own
mkdir build
cd build
cmake ..
make
If the compilation goes well, directories with binaries and static link libraries should appear.
There are several binaries in this project:
- bfm_example: This file shows how to calculate the local Bessel-Fourier moments of an image. This uses the image 'own/res/im.jpg'.
- own_example: This file shows how to calculate the features. This file uses the keypoint models in bfm_centres_M.N_K.h where M, N and K are the parameters described in the publication. This uses the image 'own/res/im.jpg'.
- own_train: This file trains the keypoint models using K-means clustering. To use this file you will have to create some directories:
- own/res/dataset -> This directory must contain a file named 'listfile.txt' containing the path of each of the images to be used for training, one path per line and no other lines are accepted.
- own/res/output -> The output of the training is a file in this directory called 'centres.xml', this is a file containing the cv::Mat of the learned centroids with the opencv tool shown here. To use these centres you will have to put them in a header file like the 'bfm_centres_M.N_K.h' and include them in your implementation.
This code has been tested in Ubuntu 16.04 and OSX 10.11
The Matlab code is only an example and may be tricky to compile, this functionality has only been tested in OSX 10.11.
Jaime Lomeli-R. (2016) University of Southampton [email protected]