Energy-based architecture for classification of publication figures

We present an implementation of the experimental and theoretical results obtained in the analysis of text and image content of biomedical publications. Particularly, we propose a novel optical recognition system using an adaptive algorithm for the classification and analysis of highly heterogeneous images in research papers. When compared with conventional algorithms, our technology substantially increases the probability of detection and classification of images buried in text or obscured by other images. We report successful testing of the new architecture using PubMed publications.