面向大数据的预期形状导向压缩感知成像的重建方法研究

Compressed sensing theory plays a beneficial role in the ability of signal acquirement. However, the efficiency and quality of signal reconstruction cannot meet an urgent demand of the fields such as remote sensing imaging, medical diagnostic imaging and calculation imaging. Different from that the traditional compressed sensing reconstruction considers the whole and detail simultaneously, the calculation process of human visual perception is to reconstruct visible geometries partitioned by color, gray scale and texture, based on the prior knowledge and the expectation. How to reconstruct signals, to reconstruct the main information with a higher quality and to improve the reconstruction efficiency according to the human visual perception mechanism are the main issues studied in this project. Firstly, in this project, the sparse representation is extended from independent and identically distributed single signal to big data group signals which have high redundancy in feature structure. Therefore, it can characterize humans' prior knowledge and express the expected information more effectively. Secondly, an innovative idea that signals can be reconstructed according to expected shape is proposed, which not only keeps the high quality reconstruction of main information, but also reduces the redundancy. Lastly, an optimization strategy based on target constraints is proposed, which can improve the reconstruction efficiency. The contributions of this project include: 1) the sparse representation of group signals with associated structural features, 2) the optimized reconstruction strategy under constraints with expected shape guidance. The research achievements of this project can provide a good theoretical and methodological foundation for information mining and value discovery for big data in the current cloud computing environment.

压缩感知理论对于提高信号获取效率和能力非常有利，但其重构效率和质量还远不能满足遥感探测成像、医学诊断影像、计算成像等领域的需求。传统压缩感知重构是同时考虑整体与细节，而人类视觉感知的计算过程是用先验知识和预期重构出由颜色、灰阶、纹理等划分的可见面几何形状。如何按人类视觉感知信息机理重构信号，并高质量反演主要信息和提高重构效率，是本课题重点攻克的目标。本课题首先将稀疏表示从独立同分布的单信号内部扩展到特征结构高度冗余的大数据中的群信号，从而更有效地刻画人们的先验知识和表达预期信息；其次是提出了按预期形状重构信号的创新思想，既保留影像中重要信息高质量重构，又可减少冗余数据产生；最后提出基于预期形状导向约束的优化策略，可保证重构效率的提高。本项目的创新点：群体信号结构特征关联稀疏表示，预期形状导向约束优化重构策略。项目的研究成果可为当前云环境下大数据中的信息挖据、价值发现提供良好的理论和方法基础

压缩感知理论对于提高信号获取效率和能力非常有利，但其重构效率和质量还远不能满足遥感探测成像、医学诊断影像、计算成像等领域的需求。传统压缩感知重构是同时考虑整体与细节，而人类视觉感知的计算过程是用先验知识和预期重构出由颜色、灰阶、纹理等划分的可见面几何形状。如何按人类视觉感知信息机理重构信号，并高质量反演主要信息和提高重构效率，是本课题重点攻克的目标。已完成研究内容包括：（1）将稀疏表示从独立同分布的单信号内部扩展到特征结构高度冗余的大数据中的群信号，从而更有效地刻画人们的先验知识和表达预期信息；（2）提出了按预期形状重构信号的创新思想，既保留影像中重要信息高质量重构，又可减少冗余数据产生；（3）提出基于预期形状导向约束的优化策略，可保证重构效率的提高；（4）本项目完成了群体信号结构特征关联稀疏表示，预期形状导向约束优化重构策略。项目的研究成果可为当前云环境下大数据中的信息挖据、价值发现提供良好的理论和方法基础。总之，研究成果达到预期效果。