Large-scale classification and retrieval for complex data

Abstract

The exponential growth of digital information production and accumulation, as well as its huge diversification, present considerable management and access challenges. At the same time, new opportunities are created by the application of analysis methods for the discovery of interesting and potentially useful patterns, latent in the data sets. In this context, recently, we have created the ReCOD Laboratory (Reasoning for Complex Data) in the Institute of Computing at UNICAMP. Aligned with the ReCOD, we present this project as a key step for its first two years organization. The term Complex Data used in this proposal cromprises the universe of unstructured, semi-structured and multimodal data, and therefore covers a wide variety of data as text, sound, image, video, geographic information, among others. Although independent research have been developed in recent years to deal with each of these types of information, right now we are living an important convergence time and, relying on on recent work on Feature Extraction, Multidimensional Indexing, Machine Learning and Fusion of Evidences we are, more than ever, able to create a coherent framework to handle these data. Many interesting questions are still open regarding these topics, conveying opportunities for valuable research contributions. We will validate the proposed work through rigorous experiments and with large databases which will allow statistically significant evaluations. Furthermore, we will show the versatility of the proposed platform in three important applications: Multimedia Forensics, Information Retrieval and Multimedia Classification, and Remote Sensing Image Recognition. This project is part of a top priority research topic both internationally and nationally. It is aligned to one of the SBC's Grand Challenges in Computing: Information Management in Large Volumes of Distributed Multimedia Data. (AU)