Thursday, 28 Mar, 2024

+91-9899775880

011-47044510

011-49075396

Motion Features for Content-Based Video Classification in Dynamic Backgrounds

International Journal of Research in Signal Processing, Computing & Communication System Design

Volume 1 Issue 2

Published: 2015
Author(s) Name: Narra. Dhanalakshmi, Y. Madhavee Latha, A. Damodaram | Author(s) Affiliation:
Locked Subscribed Available for All

Abstract

In this paper we present an innovative and accurate way of classifying videos into different genres by extracting motion features in dynamic backgrounds. Extracting motion features from the videos having dynamic backgrounds is a challenging task. For content-based video classification, first foreground information is obtained by subtracting background using Histogram of Oriented Gradients (HOG) and graph-cut methods. The combination Histogram of Oriented Gradients (HOG) and Graph Cut methods is used to remove the background to extract only foreground objects of interest. It facilitates an accurate extraction of motion features even in varying backgrounds. In order to make the algorithm less complex in dealing large volumes of video data key-frames are extracted by comparing consecutive frames based on Pearson Correlation Coefficient (PCC). It leads to efficient design of video classification algorithms in terms of processing time and energy consumption. The classification of videos will be done based on the Motion analysis of foreground objects only. For this motion features are extracted based on the temporal differencing method for characterizing the video content. Hidden Markov Model (HMM) is used as a supervised classifier because it can better learn the motion features of the video. HMM executes two different processes: Training process is used to build a model for each genre by learning the motion features and Testing process includes applying these models for classifying unknown video into one of the predefined genres. We tested our scheme on video database with some purpose based video genres such as entertainment (Sports, Cartoons, and Documentaries) and information (News). Experiments show that the proposed algorithm is capable of achieving good mean Accuracy Rate with 0.86 and also tolerable mean Error Rate with 0.375.

Keywords: Hidden Markov Model (HMM), Content-based Video Classification, HOG, Graph Cut Method, Background Subtraction

View PDF

Refund policy | Privacy policy | Copyright Information | Contact Us | Feedback © Publishingindia.com, All rights reserved