Application-Specific Evaluation of Content Based Video Retrieval Methods
F. Becker, Martin Schmücker · TUbilio (Technical University of Darmstadt) · 2005
Content-based video retrieval (CBVR) is still in its infancy: practical implementations almost never go beyond queries on textual information on scenes or complete video sequences' descriptors annotated by a human being. Automated annotation is mostly found on research-oriented systems. The original purpose of this study was to provide tools for selection and integration of descriptors in a CBVR system dedicated to specific application purposes. The objective of this bibliographical review is to present and compare various visual low-level descriptors found in scientific literature and video standards, with a real or potential interest for a CBVR system; we address different issues: either methodological (criteria for their selection and evaluation, and/or their pertinence and performance), linked to image and signal processing, or computer oriented like underlying data structures, etc.