Department of Information Technology
Information processing research lab
Since the underlying tool in many pattern recognition, clustering, and machine learning algorithms is actually a selected distance function, or more generally, a selected similarity measure, it is motivated to study similarity measures to understand what kind of bias they apply to algorithms based on them.
This project was started at summer of 2000 by Ph.D. students Ville Kyrki and Joni Kamarainen, and research assistant Jarmo Ilonen. The primary goal of the project was to gain authors knowledge on similarity measures. As a result authors introduced a new linear transform, called the neighbor-bank transform, to reduce the dimensions of the data and simultaneously increase information content of the data vectors. The neighbor-bank transform is a useful pre-processing technique for ordered histogram type of data, e.q., gray-level histograms and color spectra. See documentation for more information
Project is finished. All the goals were met.
Scientific articles produced by this project in chronological order (contact authors for reprints).
Matlab(tm) source code (for scientific and non-profitable use only)
|Function for neighbor-bank generation||newneighborbank.m|
|Script for demonstration of properties of neigbor-bank projection||demo01.m|
Data sets used in our experiments:
|1||Trees data set||Download page|
|2||Lumber data set||Download page|
|3||Bubbling data set||Download page|
|Joni Kamarainen||WWW||Responsible author, researcher, Lappeenranta University of Technology|
|Jarmo Ilonen||WWW||Research assistant, Lappeenranta University of Technology|
|Ville Kyrki||WWW||Researcher, Lappeenranta University of Technology|