t-thanh / multispectral-interest-points Goto Github PK

MULTISPECTRAL INTEREST POINTS FOR RGB-NIR IMAGE REGISTRATION
This software was used to implement and test the methods researched in the paper by D. Firmenich, M. Brown and S. Süsstrunk (EPFL, Swiss Federal Institute of Technology) as presented at ICIP 2011.


CREDITS
Written by Damien Firmenich
Project supervised by Matthew Brown

Harris is inspired by VLFeat (vlfeat.org)
SIFT is based on the paper by Lowe (2004)


INSTALLATION
Dependencies: see LIBRARIES
Python 2.5+ should be installed with the required libraries

For the C code to be used by Python, the Makefile compiles it to a shared library, please run:
    make

and run the tests with the python script:
    python tests.py


STRUCTURE
data/	data generated by the tests
lib/	external libraries used in C
media/	photos used for the tests
src/	source files


ARCHITECTURE
Python:
	- load images
	- run tests
	- write to files
	- plot

C:
	- array manipulation
	- Harris
	- DoG
	- SIFT descriptor + frames
	- registering
	- RANSAC


LIBRARIES
ctypes (interaction between Python and C)
matplotlib (plotting the data)
numpy (array manipulation)


RESOURCES
http://blogs.oregonstate.edu/hess/code/sift/
http://www.cs.unc.edu/~ccwu/siftgpu/