Data Modeling and Hybrid Query for Video Database

• Main Author: Affendey, Lilly Suriani
• Format: Thesis
• Language: English
• Published: 2006
• Subjects: Video recordings - Databases.; Information modeling.
• Online Access: http://psasir.upm.edu.my/id/eprint/5872/1/FSKTM_2006_7%20IR.pdf; http://psasir.upm.edu.my/id/eprint/5872/
• Institution: Universiti Putra Malaysia

Video data management is important since the effective use of video in multimedia applications is often impeded by the difficulty in cataloging and managing video data. Major aspects of video data management include data modelling, indexing and querying. Modelling is concerned with representing the structural properties of video as well as its content. A video data model should be expressive enough to capture several characteristics inherent to video. Depending on the underlying data model, video can be indexed by text for describing semantics or by their low-level visual features such as colour. It is not reasonable to assume that all types of multimedia data can be described sufficiently with words alone. Although query by text annotations complements query by low-level features, query formulation in existing systems is still done separately. Existing systems do not support combination of these two types of queries since there are essential differences between querying multimedia data and traditional databases. These differences cause us to consider new types of queries. The purpose of this research is to model video data that would allow users to formulate queries using hybrid query mechanism. In this research, we define a video data model that captures the hierarchical structure and contents of video. Based on this data model, we design and develop a Video Database System (VDBS). We compared query formulation using single types against a hybrid query type. Results of the hybrid query type are better than the single query types. We extend the Structured Query Language (SQL) to support video functions and design a visual query interface for supporting hybrid queries, which is a combination of exact and similarity-based queries. Our research contributions include a video data model that captures the hierarchical structure of video (sequence, scene, shot and key frame), as well as high-level concepts (object, activity, event) and low-level visual features (colour, texture, shape and location). By introducing video functions, the extended SQL supports queries on video segments, semantic as well as low-level visual features. The hybrid query formulation has allowed the combination of query by text and query by example in a single query statement. We have designed a visual query interface that would facilitate the hybrid query formulation. In addition we have proposed a video database system architecture that includes shot detection, annotation and query formulation modules. Further works consider the implementation and integration of these modules with other attributes of video data such as spatio-temporal and object motion.