Transactions on Mass-Data Analysis of Images and Signals (ISSN:1868-6451)
Volume 2- Number 1 - September 2010 - Pages 68-82
Combining Low-level Features for Improved Classification and Retrieval of Histology Images
J.C. Caicedo1,2 and E. Izquierdo2
1Universidad Nacional de Colombia, Bioingenium Research Group
2Queen Mary University of London, Multimedia and Vision Research Group
Abstract
Feature combination for image classification and indexing is an important design aspect in modern image retrieval systems. It is particularly valuable in medical applications and specially in histology applications in which different features are extracted to estimate tissue composition and architecture. This paper presents an experimental evaluation of textural features combination for histology image classification and retrieval, following a late-fusion scheme. The main focus of this evaluation is oriented to feature normalization to guarantee fair conditions for feature comparison and integration. The experimental evaluation was carried out on a collection of histology images to evaluate the feature combination strategy. Experimental results show that it is possible to improve the system performance by appropriately considering the structure and distribution of visual features. Also, it is shown that feature combination may lead to a decreased performance due to fundamental differences between image descriptors.