Tagging Enrichment Algorithm Based on Tag Semantic and Image Visual

Deng-di SUN,Mei-ling GE,Zhuan-lian DING,Bin LUO
2017-01-01
Abstract:In order to solve the problem that user-provided tags are incomplete and inaccurate,an image tagging enrichment algorithm based on tag semantic and image visual is proposed.In this approach,co-occurrence probability and WordNet are both utilized to explore latent tag correlations,for selecting candidate tags for images.Then visual features and high-level semantic are also both considered when calculating image similarity,to determine the neighborhood relationship between images.Finally based on the neighbor images and the candidate tags,a new neighbor soft-voting strategy is adopted and the tags with high voting score are reserved.Experiments taken on datase MIRFlickr-25k,systematically analyzing the influence of different semantic similarity and image similarity measurements,experimental results demonstrate that the proposed method can solve the problem of incomplete tags effectively.
What problem does this paper attempt to address?