Multimedia auto-annotation via label correlation mining. (16th April 2019)

• Record Type: Journal Article
• Authors: Tian, Feng; Shang, Fuhua; Sun, Ning
• Is Part Of: International journal of high performance computing and networking. Volume 13:Number 4(2019)
• Page Start: 427
• Page End: 435
• Publication Date: 2019-04-16
• Subjects: label correlation -- multimedia annotation -- auto-annotation -- correlation mining; High performance computing -- Periodicals; Computer networks -- Periodicals; High performance computing; Periodicals; 004.05
• Languages: English
• ISSNs: 1740-0562

Abstract

How to automatically determine the label for multimedia object is crucial for multimedia retrieval. The neighbour voting mechanism is known to be effective for multimedia object annotation. However, it estimates the relevance of a label with respect to multimedia content by labels' frequency derived from its nearest neighbours, which does not take into account the assigned label set as a whole. We propose LSLabel, a novel algorithm that achieves comparable results with label correlation mining. By incorporating the label correlation and label relevance with respect to multimedia content, the problem of assigning labels to multimedia object is formulated into a joint framework. The problem can be efficiently optimized in a heuristic manner, which allows us to incorporate a large number of feature descriptors efficiently. On two standard real world datasets, we demonstrate that LSLabel matches the current state-of-the-art.