EVSBE: Extended Visual State Binary Embedding Model for Efficient, Scalable and Fast Video Event Retrieval

Abstract

With the exponential increase of media data on the web, fast media retrieval is becoming a significant
research topic in multimedia content analysis, analysis of video content has gained growing research interest in domain
of computer vision and multimedia. In video content analysis, retrieval of event in unconstrained scenarios vital research
problem because of large scale unstructured visual information from the video descriptions. There are number of
methods and models designed for video event retrieval, but suffered from the various limitations such as scalability,
processing speed and efficiency. In this paper, the designing an efficient, scalable and fast model for video event retrieval
by considering visual approach, semantic approach and relevance feedback approach. VSBE model is designing in order
to encode the video frames which are containing the important semantic data in binary matrices. This helps to achieve
the fast event retrieval under unconstrained scenarios. The approach needs limited key frames from the training event
videos for the functioning of hash training so that complexity of computation will be less during training process.
Additionally, VSBE model applying the pairwise constraints those are generated from the visual states for stretching the
events local properties as semantic level in order ensure the accuracy. In second contribution, is extending the VSBE
model called Extended VSBE
(EVSBE) in order address the problem of end user satisfaction and out of event videos by using algorithm of log based
relevance feedback. The performance will be evaluated in terms of precision, recall, accuracy and training time