Multilayer ensemble of ELMs for image steganalysis with multiple feature sets
Punam Bedi, Veenu Bhasin · International Journal of Computational Science and Engineering · 2017
A multilayer ensemble of extreme learning machines (ELMs) for multi-class image steganalysis is proposed in this paper. The proposed ensemble consists of three levels and uses multiple feature sets extracted from images. The first two layers form sub-ensembles, one sub-ensemble for each of the feature sets. Each feature set is partitioned and used with multiple ELMs at level-1. These feature sets along with the output of the ELMs at level-1 are used by different ELMs at level-2 to classify images into multiple classes. To combine these results from sub-ensembles a stacking technique is used. Results of level-2 ELMs are used as input for the last level ELM. The fast learning process of ELM aids the speedy execution of the proposed method. Performance of the proposed method is compared with existing steganalysis methods based on individual feature sets and on 2-level ensemble. The experimental study demonstrates that the proposed method classifies images into multiple classes with higher accuracy and this has been confirmed using t-test with 99% confidence.