Fusing image representations for classification using support vector machines

TL;DR

This paper compares feature-level and classifier fusion methods for image classification, finding that classifier fusion, especially Bayes belief integration, yields superior accuracy.

Contribution

The study evaluates and compares main fusion strategies for image representations, highlighting the effectiveness of classifier fusion over feature-level fusion.

Findings

Classifier fusion outperforms feature-level fusion.

Bayes belief integration is the most effective classifier fusion method.

Experimental results demonstrate improved classification accuracy.

Abstract

In order to improve classification accuracy different image representations are usually combined. This can be done by using two different fusing schemes. In feature level fusion schemes, image representations are combined before the classification process. In classifier fusion, the decisions taken separately based on individual representations are fused to make a decision. In this paper the main methods derived for both strategies are evaluated. Our experimental results show that classifier fusion performs better. Specifically Bayes belief integration is the best performing strategy for image classification task.