zhenglab/lgmm

Registering Retinal Vessel Images from Local to Global via Multiscale and Multicycle Features

LGMM

Registering Retinal Vessel Images from Local to Global via Multiscale and Multicycle Features.

Copyright

This software is free for use in research projects. If you publish results obtained using this software, please use this citation.

@InProceedings{zheng2016registering,
author = {Zheng, Haiyong and Chang, Lin and Wei, Tengda and Qiu, Xinxin and Lin, Ping and Wang, Yangfan},
title = {Registering retinal vessel images from Local to Global via Multiscale and Multicycle features},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2016}
}

Installation

1. In Dataset/Image/, there is only one pair of example images. The index.txt presents one pair of retinal images from same eye in every row.
2. Run ./preconditioning.m first to preprocess retinal images and obtain skeleton images, the results are saved in Dataset/Skeleton/.
3. The file sdfs.cpp is known as Space-based Depth-First Search algorithm for finding the cycle structures, which should be compiled prior to use such as g++ sdfs.cpp -o sdfs in Linux and Mac OS X systems.
4. Run ./registration.m to save optimal registration result of retinal images in Results/.

Notes

1. The code is run with 64-bit Matlab R2013a on Mac OS X Yosemite, so two *.cpp files should be recompiled in other operating systems, which are in PreProcessing/mex.
2. The legacy flag of unique and intersect functions should be removed in Matlab R2012b and prior releases in your code, and the two functions are applied in export_featuremat.m, export_loop.m and find_loop.m files.