Efficient cross-modal search through deep binary hashing and quantization

摘要

To provide a new method for cross-modal retrieval via deep binary hashing and quantization.SOLUTION: In a training phase, a system simultaneously learns to generate feature vectors, binary codes, and quantization codes that preserve the semantic similarity correlations in the original data for data across multiple modalities. In a prediction phase, the system retrieves an item that is semantically similar to a query having a different modality from items in a database. In order to identify items closest in semantic meaning to the query, the system first narrows a database search space based on binary hash code distances between each item and the query. The system then measures quantization distance between the query and the database item in the smaller search space. The system identifies an item having the closest quantization distance to the query as the closest semantic match to the query.SELECTED DRAWING: Figure 1