Summary
Smart Info-Media Systems in Asia
2019
Session Number:RS3
Session:
Number:RS3-7
A Graph-based Video Visual Reranking Method via Heterogenous Graph Analysis
Soh Yoshida, Mitsuji Muneyasu,
pp.161-166
Publication Date:2019/9/4
Online ISSN:2188-5079
Summary:
This paper addresses the problem of analyzing topics, included in social videos, for improving the performance of video retrieval. Unlike previous works which only focus on an individual video visual aspect, the proposed method leverages the "mutual reinforcement" of heterogeneous objects such as text tags and users. In order to represent multiple types of relationships between each heterogeneous object, the proposed method constructs three subgraphs: a user-tag graph, a video-video graph, and a videotag graph. We combine the three types of graphs to obtain the heterogeneous graph. Then the extraction of latent features, i.e., topics, becomes feasible by applying graph-based soft clustering to the heterogeneous graph. By estimating the membership of each grouped cluster for each video, the proposed method defines a new video similarity measure. Since the understanding of video content is enhanced by taking advantage of latent features obtained from different types of data, that complement each other, the proposed method can improve the performance of video visual reranking. We conduct experiments on the YouTube-8M dataset, and the results show that our reranking approach is effective and efficient.