低秩稀疏约束的图像大数据感知哈希算法研究

Perceptual image hashing technology has wide application in image retrieval, image authentication, and so on. However, existing perceptual image hashing algorithms didn’t consider the big data characteristics of image data in network environment very well, and their real-time, robustness and discrimination requirements need to be improved. So, they cannot be effectively applied to big data image processing and computation. From the point of view of sparsity of big data image, a big data image perceptual hashing research scheme based on low-rank and sparsity is proposed in this project, which uses low-rank and sparsity theory to fully exploit value sparsity (low density) and visual perception sparsity of big data image. The basic theory and key technology of big data image perceptual hashing will be explored in this project. Firstly, big data sampling method based on low-rank decomposition and low-rank and sparsity model for big data image are studied. Secondly, low-rank feature extraction strategies and basic theory and methods of big data image perceptual hashing algorithms with low-rank and sparsity constraints are studied. Thirdly, oriented at image retrieval, image authentication and image deduplication, the experiment platform of big data image perceptual hashing algorithms based on low-rank and sparsity are established. Finally, sparse construction is used to develop the performance benchmark of low–rank based big data image perceptual hashing algorithms. The study will develop new theories and methods for big data image perceptual hashing technology and promote the steady and consistent development of big data image perceptual hashing and big data image computation application.

图像感知哈希技术在图像检索、认证等领域应用广泛。现有图像感知哈希算法缺乏对网络环境图像大数据内在特性的深入分析与利用，算法实时性、鲁棒性与区分性有待提高，难以适应于图像大数据处理应用。本项目从图像大数据数据冗余性与价值稀疏特性出发，提出基于低秩稀疏优化的图像大数据感知哈希算法研究方案，利用低秩稀疏优化理论，充分挖掘图像大数据价值稀疏性与视觉感知稀疏性，探索图像大数据感知哈希理论与关键技术。具体研究内容包括基于低秩分解的图像大数据稀疏采样机制、图像大数据低秩稀疏模型；研究图像大数据低秩稀疏特征提取，研究基于低秩稀疏表示的图像大数据感知哈希算法理论与方法；面向图像检索、认证与消冗应用构建图像大数据感知哈希算法原型验证平台，研究基于稀疏重构的图像感知哈希算法性能评测准则。项目研究将为图像大数据感知哈希算法提供新的理论与方法，推动图像大数据感知哈希技术及图像大数据应用的健康持续发展。

图像感知哈希技术在图像检索、认证等领域具有广泛的应用。针对通常的图像感知哈希算法未能充分挖掘图像大数据特性，本项目从图像大数据数据冗余性与价值稀疏特性出发，提出基于低秩表示的图像大数据感知哈希算法研究方案，项目通过充分挖掘利用图像大数据稀疏性，建立基于低秩表示的图像大数据采样机制、图像大数据低秩特征计算模型，设计了基于低秩表示LBP特征的图像大数据感知哈希算法及及面向加密域的图像感知哈希方法；从安全性、鲁棒性、唯一性、检索准确性等多方面验证测试了基于低秩特征的图像感知哈希算法。研究成果丰富了图像感知哈希生成方法，可应用于云计算环境图像检索与图像内容认证，推动图像大数据应用的健康持续发展。