Bayesian Network-based Active Learning for Ontology-based Video Retrieval using Hidden Markov Logic Model with Edge Histogram Descriptor

A large amount of data is collected, analysed, and used on a regular basis by a large number of people for web pages, books, journals, and files which make up a significant portion of the data on the Internet. The burden on significant content-based retrieval and other applications is increasing as the volume of surveillance video data grows. Therefore to reduce the burden during retrieval and to solve the problem of video surveillance an ontology based video retrieval by feature extraction and classification techniques is proposed. In the Initial step the examined form of description files in OWL (Web Ontology Language is fed as input. Then the video pre-processed and segmented before extracting the features. The features of video has been extracted using "Hidden Markov Logic Model With Edge Histogram Descriptor (HMLM_EHD)" in which the spatio based features are extracted with Shape-based features that define object shapes in image is extracted from object contours or regions. After the extraction of features, classification is done using Bayesian network based active learning. The retrieval subsystem evaluates the provided query, performs similarity matching operations (which can be done using the correlation Distance Algorithm), and finally provides the result to the end user for these categorised video frames. The experimental results shows retrieved video frames and their parametric analysis in terms of accuracy of 94%, precision of 85%, recall of 71 % and F-1 score of 73%.