The identification of objects in natural images and the semantic interpretation of the image is an important topic of research. For a collection of images, image analysis typically involves a number of steps that employ predetermined tools. The choice of i

Region based classifier selection in image understanding

SAMEER SINGH, Member IEEE

MANEESHA SINGH, Student Member, IEEE

Department of Computer Science

University of Exeter Exeter EX4 4PT United Kingdom

Tel: +44-1392-264053 Fax: +44-1392-264067

Email: {s.singh, m.singh}@ex.ac.uk

ABSTRACT

The identification of objects in natural images and the semantic interpretation of the image is an important topic of research. For a collection of images, image analysis typically involves a number of steps that employ predetermined tools. The choice of image segmentation and texture analysis algorithms and their parameters for processing more than one image is based on the a priori experience of the researcher and what in their opinion seems to work the best for a given application. It has been recently shown that the optimisation of image processing tools on a per image basis is likely to lead to high quality classification results for object identification. In this paper we develop a ‘bank of classifiers’ approach to image object recognition and evaluate both selective and classifier combination approaches against a baseline approach that works with a single classifier.

Index Terms: Scene analysis, benchmark, object recognition, texture analysis, image segmentation, classifier combination

1. Motivation

It is well-known that no single image analysis tool is best suited for all images and the performance evaluation of computer vision algorithms remains an important research topic [27]. The selection of image processing algorithms based on statistical properties of the image has received some attention