PROJECT TITLE :

Semantic Neighbor Graph Hashing for Multimodal Retrieval - 2018

ABSTRACT:

Hashing strategies are widely used for approximate nearest neighbor search in recent years due to its computational and storage effectiveness. Most existing multimodal hashing methods strive to preserve the similarity relationship based mostly on either metric distances or semantic labels in an exceedingly procrustean way, while ignoring the intra-category and inter-category variations inherent in the metric space. During this Project, we propose a completely unique multimodal hashing methodology, termed as semantic neighbor graph hashing (SNGH), which aims to preserve the fine-grained similarity metric primarily based on the semantic graph that's constructed by jointly pursuing the semantic supervision and therefore the native neighborhood structure. Specifically, the semantic graph is constructed to capture the local similarity structure for the image modality and therefore the text modality, respectively. Furthermore, we outline a function based mostly on the local similarity in particular to adaptively calculate multi-level similarities by encoding the intra-class and inter-class variations. When getting the unified hash codes, the logistic regression with kernel trick is utilized to be told view-specific hash functions independently for every modality. In depth experiments are conducted on four widely used multimodal knowledge sets. The experimental results demonstrate the prevalence of the proposed SNGH methodology compared with the state-of-the-art multimodal hashing strategies.