FILE STRUCTURES FOR ICONIC PICTORIAL RETRIEVALS

Abstract

With the rapid growth of multi-media systems and multi-media technology, information systems no longer deal with information that is purely in text form but also with pictures and graphics to complement the traditional way of presenting information. In recent years, the so-called content-based image database systems are introduced. New methods to organize the image data based on the contents of the image are brought forth. But the main reason behind the slackening of advance in image database system is the limitation of image analysis. This thesis attempts to introduce the concept of Animaled Pictorial Database (APD) which provides more flexibility and functionality to the database designers and users. Designing APDs starts with the basic level of icons, and builds pictures up hierarchically. Therefore no feature extraction or image analysis are needed. In particular, we propose the Pictorial Iconic Access Relation (PIAR) to facilitate similarity retrievals from pictorial databases. Each entry in PIAR consists of a Picture ID, an Icon ID that represents one of the symbolic objects in the picture, the coordinates of the left-bottom and right-top corners of the rectangle that enclose the symbolic object. This form of implicit representation of spatial relationship between symbolic objects saves storage space and also solves the problem of false match ambiguity. The advantages of PIAR, such as less storage cost, ease of update and no ambiguity problem in similarity retrievals are discussed. Additionally, PIAR provides a unifying approach to iconic indexing for two, three or higher dimensional scenes. To support our proposed PIAR, a Content-Based Iconic Indexing Scheme (CBIIS) is also proposed. Moreover the main thrust of this thesis is the design and implementation of presentation subsystem and storage subsystem of pictorial objects that supports CBIIS to process different types of queries that are frequently issued to an APD. Above all, the thesis also deals with a performance study that will demonstrate that our proposed PIAR is a time-efficient retrieval method .