PYTHON WRAPPER FOR ORSA

OVERVIEW

This source code is a fork of an IPOL publication. Please cite http://www.ipol.im/pub/art/2012/mmm-oh/ if you use this code as part of your research. It provides an additional python wrapper to use the code natively in python.

COMPILATION

The code is compilable on Unix/Linux and on Mac OS (requires a simple manual fix though!).

Compilation: requires the cmake and make programs.

Dependencies: Requires libpng, libtiff and libjpeg.

Configure and compile the source code using cmake and make. It is recommended that you create a folder for building:

UNIX/LINUX/MAC:

$ mkdir build; cd build
$ cmake ../src
$ make

/!\ If you're on mac, the libraries are compiled as .dlib instead of .so. You'll need to change the extension for the provided python scripts to work out of the box (this will be fixed later)

$ mv demo/liborsa.dlib demo/liborsa.so
$ mv demo/libworsa.dlib demo/libworsa.so

USAGE

The wrapper can be accessed by importing orsa in python:

from orsa import orsa

This library provides a single function orsa.estimate_homography_py(im1, w1, h1, c1, im2, w2, h2, c2, precision, siftRatio, H, inl, outl, im1w, im2w, mosaic):

• im1: Numpy array corresponding to the first image. It has a shape of (c1, h1, w1) and be of type np.float32.
• im2: Numpy array corresponding to the first image. It has a shape of (c2, h2, w2) and be of type np.float32.
• precision: Precision of the processing. Recommended value is 0.
• siftRatio: Sift ratio used during the matching part. Recommended value is 0.6.
• H: Numpy array that will contain the estimated homography (it needs to be of shape 9 and of type np.float32).
• inl and outl: Numpy arrays that will contain a visual represensation of the inliers for inl and the outliers for outl. They both need to be of shape (3, max(h1,h2) * max(w1,w2)/(w1+w2), max(w1,w2)) and be of type np.float32. It is not be computed when set to NULL.
• im1w, im2w, mosaic: Numpy arrays that will contain respectively the warped im1, the warped im2 and a mosaic made from the two. They all need to be of shape (3, max(h1,h2), max(w1,w2)) and be of type np.float32. It is not be computed when set to NULL.

A test code main.py is also provided to show how this can be used. Dependencies for this test code can be installed using the provided requirements file requirements.txt.