Introduction

This repository contains a python implementation of the Region Based Easy Path Wavelet Transform [1]. The latest version of this software can always be found on github: https://github.com/nareto/rbepwt

Dependencies

The following python packages are used:

numpy
scipy
pywt
skimage

External Files

In order to compute the VSI and HaarPSI indexes the files VSI.m [2] and HaarPSI.m [3] respectively are needed, as long as a working octave installation and the python package oct2py.

Example Usage

Create an Image instance and load an image file:

import rbepwt
rbimg = rbepwt.Image()
rbimg.read('/path/to/cameraman.png')

Segment it using the Felzenszwalb-Huttenlocher method and view the segmentation:

rbimg.segment(method='felzenszwalb',scale=200,sigma=2,min_size=10)
rbimg.show_segmentation()

Encode it using the RBEPWT method with the easypath path-finding procedure, 16 levels and the CDF 9/7 wavelet (bior4.4 in pywt); keep only the greatest 512 coefficients and decode:

rbimg.encode_rbepwt(16,'bior4.4',path_type='easypath',euclidean_distance=True)
rbimg.threshold_coefs(512)
rbimg.decode_rbepwt()

View the result:

rbimg.show_decoded()

[1] Budinich, R. (2017b). A region-based easy-path wavelet transform for sparse image representation. International Journal of Wavelets, Multiresolution and Information Processing, 15(05):1750045.

[2] http://sse.tongji.edu.cn/linzhang/IQA/VSI/Files/VSI.m

[3] http://www.math.uni-bremen.de/~reisenho/software/HaarPSI.m