waveblend - Focus enhancing by image fusion using the dualtree complex wavelet transform

by Sebastian Nowozin (nowozin@cs.tu-berlin.de)

Introduction

Waveblend is a tool implementing an image fusion algorithm. It can combine a number of input pictures having all the same size into one output picture. While this could be useful for a number of purposes, waveblend focuses on only one: to enhance the focus of the combined image.

Focus enhancing can be explained roughly like this: Take a number of input images, all taken from the same point of view with different depth-of-focus (that is, they are sharp at different depths). All the images are combined into one image, which spans the whole depth-of-focus of all images, so that the resulting image is "sharp" everywhere.

This has a number of applications: in microscopy, astronomy and photography you can improve the results of a sample image by combining a number of sample images. A description of the techniques used is given below.

Screenshots

Everybody likes screenshots, so here are some screenshots of waveblend. Although waveblend is a command line program, the distribution has two debugging output windows you can view:

This is the blending stage debugging output you get if you enable the "--debug" option. First three input images are shown, then the grayscale fused image (already in focus) is displayed. From this image, a topology map is reconstructed, which is shown underneath. Finally, the fused color image is constructed from the topology image and the input images, which is shown as the last picture:

Fusioning of three microscopy sample
pictures to enhance the focus

Another view is not shown here. When developing the complex dualtree wavelet transform, I wrote a program to display the wavelet coefficients and wavelets. As my implementation is modelled after the Matlab implementation (referenced below), the debug output is similar. You can see the individual wavelet stages by running the "TestWavelet.exe" program included in the distribution.

Technology

The program uses the complex dual-tree 2D discrete wavelet transform as invented by Nick Kingsbury and detailed in his paper "Image Processing with Complex Wavelets". My implementation is based on the Matlab implementation done by Shihua Cai, Keyong Li, Farras Abdelnour and Prof. Ivan Selesnick, which is available at the Brooklyn Poly Wavelet Software website, which I highly recommend.

The focus enhancing algorithm used is detailed in the paper "Extended Depth-of-Focus for Multi-Channel Microscopy Images: A Complex Wavelet Approach" by B. Forster, D. Van De Ville, J. Berent, D. Sage and M. Unser.

Implementation

The whole program was developed under Debian GNU/Linux using Mono, a free implementation of the Microsoft .NET framework and compilers. The source is written completely in C#.

Requirements
I have not tested the program on Windows, but in principle it should run just fine. You might want to try the new combined Mono and Gtk# installer Novell has produced:
Combined Windows Mono 1.0 and Gtk# 1.0 installer (19mb)

Download

Changes

VersionChanges
0.1.0 Initial release.

License

The whole program is opensource and free software. It is released under the conditions of the GNU General Public License. The license is included in the distribution in the 'LICENSE' file. Please read and acknowledge this license before using this program.

Usage and Documentation

The documentation is available in the tarball in form of a README file and a manpage. Here you can find the latest documentation as PDF and in text format:
ProgramDocumentation
waveblendImage fusion tool: waveblend.pdf, waveblend.txt

Thanks

I thank the following persons for their helpful advice, implementation of algorithms or helpful texts:

Feedback

I am curious about user-, developer- and mathematicans feedback, so please mail me your thoughts to nowozin@cs.tu-berlin.de.

last update: Monday, 12 Jul 2004