LAUSR

Abstract Machine learning is constantly proving its capabilities by achieving exceptional results in recognition and classification tasks. Image content recognition has been addressed by Bags of Visual Words coupled with a classification algorithm as well as convolutional neural networks. In this work, we question the applicability of these approaches individually, while proposing two novel hybrids that work in a cooperative and synergistic way in order to achieve better content recognition performances. We focus on a real-world scenario that involves the on-demand digital content enrichment of a museum-visit experience by exploiting mobile devices. The Folklore Museum of Xanthi (Greece) has been selected as our case study. A new benchmark image dataset has been created based on the museum exhibits and used to train our recognition approaches and to quantify their performance under independent, cooperative, and synergistic operational schemata. The experiments reveal that our hybrid approaches improve the recognition performance while composing a robust framework for applications related to cultural thesaurus interaction and more specifically to the on-demand digital content enrichment with minimum infrastructure requirements.