Title

University Of Central Florida At Trecvid 2007 Semantic Video Classification And Automatic Search

Abstract

In this paper, we describe our approaches and experiments in semantic video classification (high-level features extraction) and fully automatic topic search tasks of TRECVID 2007. We designed a unified high-level features extraction framework. Two types of discriminative low level features, Spatial Pyramid Edge/Color Histograms and Bag of Visterms, are extracted from the key-frames of the shots. Then the SVM classifiers with RBF kernel are used for classification. The final classification results are produced by fusing and combining these classifiers. The experiment results show that the combined classifiers substantially improved the performance over the individual feature based classifier. In fully automatic topic search task, we mostly focus on the video retrieval using the visual content through the high level features detectors. The main challenge in this task is mapping queries to the high level features. A novel earth mover's distance (EMD) based relevance procedure that finds the similarity between queries and videos through the high level features and semantic word similarity measures.

Publication Date

1-1-2007

Publication Title

2007 TREC Video Retrieval Evaluation Notebook Papers

Number of Pages

-

Document Type

Article; Proceedings Paper

Personal Identifier

scopus

Socpus ID

84905178033 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/84905178033

This document is currently not available here.

Share

COinS