Nonstructured information retrieval based on Tolerant Granular Space Model

Granular computing is an emerging conceptual and computing paradigm of information processing. The basic idea of granular computing is the using of granulars during problem solving. A tolerance granular space model (TGSM) is proposed by intelligence science laboratory. From the perspective of philosophy, the basic idea of the model is based on the human ability, that is, people can abstract or synthetize the knowledge and data relating with special tasks to different degrees or sizes granules, and accomplish the tasks with the helps of the granules and relations among them. From the perspective of technology and theory, TGSM written as 4-tuple (OS, TR, FG, NTC), consists of four parts: object set system OS, tolerance relation system TR, transformation function FG and nested tolerance covering system NTC. The main features of the model focus on the definition of granules and the problem solving methods with the help of the hierachical and nested structure of tolerance granular spaces. Nonstructured Information comes from audio, video, image which no inherent segmentation in basic semantic units, such as characters, words, and sentences. The information retrieval should make the automatic semantic generation through feature binding. We have proposed a novel computational model, Bayesian linking field model, for feature binding in human perception, by combining the idea of noisy neuron model, Bayesian method, linking field network and competitive mechanism. In this talk we will explore how to apply tolerance granular space model for nonstructured multimedia information extraction and retrieval, particular focussing on video streams.