Scalable Cross-Media Hashing for Searching Heterogeneous Data Sources

该项目旨在使用新的索引和搜索方法来实现多媒体数据的价值。多媒体数据的日益增长的可用性和异构性要求采用新的方法来搜索跨不同媒体类型和数据源的信息。该项目旨在通过解决异构性和可扩展性问题，从各种媒体源收集的数十亿个多媒体对象中进行准确和实时的跨媒体搜索。预期的结果包括：可扩展的跨媒体哈希技术，以捕获异质数据中存在的隐含相关性，并将高维特征嵌入短二进制码；新的二进制码索引和排序方案，以进一步提高搜索速度和质量；以及大规模跨媒体系统，以评估方法并展示其实用价值。

This project aims to use novel indexing and search approaches to realise the value of multimedia data. The ever-increasing availability and heterogeneity of multimedia data are calling for new approaches to search information across different media types and data sources. This project aims to enable accurate and real-time cross-media searches from billions of multimedia objects collected from various media sources by tackling the heterogeneity and scalability issues. Expected outcomes include scalable cross-media hashing techniques to capture implicit correlations existing in heterogeneous data and embed high-dimensional features into short binary codes; new binary code indexing and ranking schemes to further improve search speed and quality; and a large-scale cross-media system to evaluate methods and demonstrate the practical value.