| Summary: | Indexing video collection with a large set of video concepts allows us to employ a query-by-concept paradigm. Through concept detection, the task is to detect in the shot- segmented video data the presence of a set of pre-defined semantic concepts. Columbia University provided 374 detectors based on color, texture and edge features, and their unsupervised classifier fusion that can be utilized for concepts training purposes. We had successfully implemented a Concept-based Video retrieval System (CBVRS) to support query-by-concept. However, one of the main challenges is to reduce the training time of concepts to support the video concept detection. This paper presents an alternative architecture to overcome the issue. The proposed CBVRS framework, which consists of three main modules i.e. pre-processing, video analysis, and annotation module, shall utilize the Hadoop MapReduce framework for fast computation of the concept detection.
|
|---|