基于数据降维和压缩感知的图像哈希理论与方法

项目名称： 基于数据降维和压缩感知的图像哈希理论与方法

项目编号： No.61300109

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 唐振军

作者单位： 广西师范大学

项目金额： 23万元

中文摘要： 图像哈希是图像处理与信息安全学科交叉的新兴前沿研究课题，可用于内容认证、质量评价、图像检索等，具有稳健性、唯一性和篡改敏感性，即，对正常处理稳健，不同图像有不同哈希，发生篡改时，哈希有重要改变。本项目根据数据降维和压缩感知理论，以特征提取和特征压缩编码为切入点，研究图像哈希的新理论和新方法，主要内容包括：研究在不同颜色空间中提取兼顾亮度、颜色和纹理等信息的稳健视觉特征；研究鲁棒视觉特征点提取、Radon投影分割、环形分割等技术，建立抗旋转变换的特征参数；运用非负矩阵分解、局部线性嵌入等降维算法来压缩特征，研究降维数据的内在规律，提出高效特征压缩编码算法；研究最优测量矩阵设计，探索测量值的压缩编码技术，建立基于压缩感知的特征压缩编码理论与方法；面向内容认证和半参考质量评价，设计图像哈希新算法；开发图像哈希系统。项目成果对我国的军事、工业和商业等掌握新的数字媒体内容安全理论和技术具有重要意义。

中文关键词： 图像哈希；压缩感知；数据降维；环形分割；局部线性嵌入

英文摘要： Image hashing is an emerging topic at the forefront of the crossover field between image processing and information security, and finds many applications such as content authentication, quality assessment and image retrieval. Its most important properties are robustness, discrimination and sensitivity to malicious tamper, which mean that image hashing is robust against normal digital processing, different images have different hashes and hash should significantly be changed once tampering operation occurs. This project is to develop new theories and methods of image hashing based on the theories of data dimensionality reduction and compressive sensing (CS). The research works mainly focus on feature extraction and feature compression. The detailed works are as follows. We will extract robust visual features about luminance, color and texture from different color spaces. We will develop efficient techniques, such as robust visual points, Radon projection based division and ring partition, to obtain robust features resilient to rotation transform. We will exploit algorithms of data dimensionality reduction, such as non-negative matrix factorization and locally linear embedding, to compress image features, investigate the inherent laws of compressed data, and then propose efficient data compression methods. We will

英文关键词： Image hash；Compressive sensing；Data dimensionality reduction；Ring partition；Locally linear embedding (LLE)