Full text
| |
| Author(s): |
Jurandy Gomes de Almeida Junior
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2011-11-28 |
| Examining board members: |
Ricardo da Silva Torres;
Mario Antonio do Nascimento;
Herman Martins Gomes;
Hélio Pedrini;
Edson Borin
|
| Advisor: | Ricardo da Silva Torres |
| Abstract | |
Recent advances in the technology have enabled the increase of the availability of video data, creating large digital video collections. This has spurred great interest in systems that are able to manage those data in a efficient way. Making efficient use of video information requires the development of powerful tools to extract high-level semantics from low-level features of the video content. Due to the complexity of the video material, there are five main challenges in designing such systems: (1) to divide the video stream into manageable segments according to its organization structure; (2) to implement algorithms for encoding the low-level features of each video segment into feature vectors; (3) to develop similarity measures for comparing these segments by using their feature vectors; (4) to quickly answer similarity queries over a huge amount of video sequences; and (5) to present the list of results in a user-friendly way. Numerous techniques have been proposed to support such requirements. Most of existing works involve algorithms and methods which are computationally expensive, in terms of both time and space, limiting their application to the academic world and/or big companies. Contrary to this trend, the market has shown a growing demand for mobile and embedded devices. In this scenario, it is imperative the development of techniques so effective as efficient in order to allow more people have access to modern technologies. In this context, this work presents five novel approaches for the analysis, indexing, and retrieval of digital videos. All these contributions are combined to create a computationally fast system for content-based video management, which is able to achieve a quality level similar, or even superior, to current solutions (AU) | |