Indexing an intelligent video database using evolutionary control

In this paper we present the implementation of an intelligent video database using evolutionary control. By using automatic video indexing techniques, the retrieval of video segments can be performed using free natural language queries. Retrieval of video segments from a database for editing and viewing is becoming an important topic in video processing. A cinematic movie consists of video segments, which are semantically related. Current approach to video retrieval emphasize on the low level semantics such as colour and textures of the video segments. However, it is difficult for the users to formulate queries in terms of these low level features. Associated with each video segment in a movie there are video scripts. Each video script contains descriptions about the content of the video and the subtitles for the video segment. Using a database of video segments with associated textual information, it is possible to provide information for video retrieval using free natural language texts. Fully automated indexing and query processing is a key problem in text-based video retrieval. To solve the associated problems, we have implemented an Automatic Video Indexing System (AVIS) using information retrieval and machine learning techniques. The system was tested using the original movie scripts from "STAR WAR - return of the JEDI" (139 movie segments) and "Star Wars - A New hope" (476 movie segments). We have formally evaluated the system using formal precision and recall measures with a fully automated indexing system. The system is able to achieve good precision-recall values. The results show that information retrieval and machine learning techniques can be applied to video information systems effectively.