暗数据是用户在开展正常业务活动期间搜集、处理、存储，但通常无法被挖掘或理解的数据。根据合规管理监督委员会的研究显示，在企业存储的数据中，69%的数据都以暗数据形式存在。暗数据占据大量的存储资源，在可利用性和安全性等方面存在诸多问题。本项目以图像暗数据为研究对象，针对其规模庞大难以挖掘利用的问题开展研究，内容包括：首先，结合深度学习、哈希技术和图嵌入技术研究具有泛化能力的内容语义哈希技术；然后，结合图排序技术和检索技术研究基于内容语义的暗数据点亮新方法；最后，结合图谱化元数据管理技术，在对象文件系统层面研究支持暗数据点亮的存储新技术。本项目创新性地提出从内容语义出发理解并点亮暗数据，并以元数据管理方式抑制暗数据的产生，促进深度学习、哈希技术与存储技术的交叉融合，为暗数据研究探索一条新的途径。

Dark data is defined as the information assets that can be easily collected and stored, but generally fail to use for data analytics and mining. According to research by the compliance management oversight board, 69 percent of the data stored by companies is in the form of dark data. The dark data occupies large storage space and poses many problems of availability and security. Faced with problems of dark data which is large and hard to analyze, this project researches new technology of lightening for image dark data by deep learning, hash technology, multi-media retrieval technology and graph ranking technology. First, we research the semantic hash method with generation ability. Then, we research new lightening method by hash method and graph ranking method. At last, Based on our new lightening method, we research new storage technology by the metadata management to reduce the generation of dark data on the object file system. This project innovatively proposes to understand and lighten dark data based on content semantics, and suppress the generation of dark data by metadata management. It can promote the fusion of deep learning, hash technology and storage technology, and explore a new way for dark data research.