Nearest Duplicate Video Retrieval with Multiple Features using Temporal Correlation

Shital Bhirud, S Desai
unpublished
Online video database is increasing exponentially; a technique is needed for the Near-Duplicate Video Retrieval (NVDR). There are many programs for this strategy, in trademark protection, video labelling, video recording tracking etc. Though, lot of methods has been developed for NDVR over the past several years. To represent the video single feature is not sufficient as fast increase in video manipulation techniques. May other methods use multiple features but they were not able to face
more » ... able to face extensive quantity of videos. Thus, these methods were having less usage in real life programs. Our method is inspired by and improves upon recent work on Multiple Feature Hashing (MFH) designed to used multiple features to interpret the image instead of single feature may not represent the image exactly. Our Evaluation of the system shows that our proposed method substantially improves upon previously proposed methods in time efficiency, accuracy and scalability. To achieve the same or more efficient search quality temporal correlation used same hashing technique as basic MFH method while reducing number of keyframes by time between them. And object detection provides more accuracy.
fatcat:2hfptrgvjzbbxo4sjeit2rqjdq