Visualizing Huge Image Databases by Formal Concept Analysis

Based on formal concept analysis we propose a novel lattice visualization system for huge image databases as a realization of the important paradigm of human-centered information processing based on granular computing. From a given cross table of objects (images) and attributes (image features) the proposed system first constructs a concept lattice. Then the Hasse diagram of this lattice is visualized. The information granules in the proposed system correspond to the elements of the concept lattice. All the important components of granular computing are shown to be present in the proposed system, such as: abstraction of data, derivation of knowledge and empirical verification of the abstraction. Since formal concept analysis generates an order relation, we obtain a hierarchical structure of concepts. This structure is shown to be also strongly related to the granular computing, since this is how the lattice visualization system implements the zoom in and zoom out capability of granular computing systems. Using the proposed system, a user can freely analyze the perspective and detailed structure of a large image database in the setting of granular computing. Furthermore, through an interaction function, the potential user can adjust the quantization of features, being able in this way, to select the attributes which allow him to obtain a suitable concept lattice. Therefore, the proposed system can be regarded as a promising human-centric information processing algorithm, based on granular computing.