Multi-feature fusion in image clustering using ant-inspired methods

Clustering of visual data is necessary for its effective organisation, summarisation and retrieval. In this paper, the appropriateness of biologically inspired models to tackle this problem is discussed and suitable strategies to solve this specific image processing task are derived. The proposed techniques are inspired by the optimal movements of ants and their biologically optimised colony behaviour. In the first proposal, the problem of multi-feature fusion using relevant, yet different, discriminative low-level features is tackled by Ant Colony Optimisation and its learning mechanism. In the second proposal, another metaheuristic model is applied. Here, the ability of ants to build live structures with their bodies is used in order to discover, in a distributed and unsupervised way, a tree-structured organisation of images. Finally, the proposed techniques are comprehensively evaluated and selected representative results are reported.