An Adaptive and Dynamic Dimensionality Reduction Method for Efficient Retrieval of Videos

Abstract

In today’s multimedia age, Content Based Video Retrieval (CBVR) is a trending area and lot of research is being carried out in Video Surveillance, Big Data analysis and multimedia applications. Usage of multimedia data is becoming very common in day today life, content based video retrieval provides an effective mechanism for maintaining, managing and retrieving large number of videos efficiently as per user’s interest. The advancement in the technology is evolving and multimedia applications are gaining more importance so the performance of the CBVR system need to be high and accurate to fulfill user demands. The proposed paper focuses on surveying the different techniques for feature extraction and similarity computation for retrieving relevant videos. Feature extraction can be done using different techniques such as shot boundary detection, based on histogram, PCA Shift, Gist and SURF (Speeded up Robust Features) and Quadratic Equation are used for feature extraction and similarity computation. In CBVR technique the videos are retrieved from the large databases based on the given Input Query as an image, on processing this query the features from videos and the query image are extracted mainly the color, texture and shape. Once these features are extracted then a similarity between query and the videos is computed, ranking will be done based on the similarity score. The results interpret that SURF technique provides better results as compared to other techniques and the system has a retrieval performance of more than 70%.