A classification based framework for concept summarization

Dhruv Kumar Mahajan, Sundararajan Sellamanickam, Subhajit Sanyal, Amit Madaan

Abstract: In this paper we propose a novel classification based framework for finding a small number of images summarizing a concept. Our method exploits metadata information available with the images to get the category information using Latent Dirichlet Allocation. We modify the import vector machine formulation based on kernel logistic regression to solve the underlying classification problem. We show that the import vectors provide a good summary satisfying important properties such as coverage, diversity and balance. Furthermore, the framework allows users to specify desired distributions over category, time etc, that a summary should satisfy. Experimental results show that the proposed method performs better than state-of-the-art summarization methods in terms of satisfying important visual and semantic properties.