欠采样相关图像集重建理论与算法

The Compressed Sensing (CS) theory, which permits a sub-sampling rate much lower than Nyquist sampling theory, reduces the difficulty of acquiring ultrahigh resolution signal. The CS theory has recently received much attention in application fields such as wireless video surveillance, wireless multimedia sensor networks, dynamic magnetic resonance medical imaging, etc. In this kind of applications, the reconstruction of sub-sampled correlated image sets is an important issue. In order to achieve the optimal reconstruction results, this project studies how to fully make use of the priori information within correlated image sets. The main research contents are listed as follows: 1) Propose efficient image reconstruction models which jointly take the priori structure information of an image and the sparsity of the related residual image into consideration. 2) Design efficient methods to solve the proposed image reconstruction models. The algorithms should be highly robust and has rapid convergence. 3) Extend the proposed image reconstruction models to support color version of correlated image sets by considering the correlation among different color channels, so that the quality of reconstruction images can be further improved. 4) Present a low complex “prediction-reconstruction” structure to help the proposed models reconstruct the whole image sets. We believe that the research of this project can offer theory reference and technical support for the application of CS theory in the filed of multimedia information acquisition.

压缩感知理论能够以远低于香农采样定理的速率进行欠采样，降低了超高分辨率信号获取的难度。该理论在无线视频监控、无线多媒体传感器网络、动态核磁共振医学成像等领域的应用都得到了广泛的关注。在上述应用场合中，欠采样相关图像集的重建问题至关重要。针对该问题，本项目研究如何充分应用相关图像集内的先验信息，以获取最优的重建图像。主要研究内容为：1) 同时考虑待重建图像的先验结构特性及残差图像的稀疏性，建立性能优异的图像重建模型；2) 分别研究所提出的图像重建模型相应的高效求解算法，算法鲁棒性强、收敛迅速；3) 考虑不同色彩通道间的相关性，将所提出的重建模型扩展至彩色相关图像集重建问题，进一步提升图像重建质量；4) 提出复杂度较低的“预测-重建”结构辅助前述模型完成整个欠采样图像集的重建。本项目的研究，可以为压缩感知在多媒体信息采集领域中的应用提供理论参考和技术支持。

压缩感知理论能够以远低于香农采样定理的速率进行欠采样，降低了超高分辨率信号获取的难度。该理论在无线视频监控、无线多媒体传感器网络、动态核磁共振医学成像等领域的应用都得到了广泛的关注。在上述应用场合中，欠采样相关图像集的重建问题至关重要。针对该问题，本项目研究如何充分应用相关图像集内的先验信息，以获取最优的重建图像。主要研究内容为：1) 同时考虑待重建图像的先验结构特性及残差图像的稀疏性，建立性能优异的图像重建模型；2) 分别研究所提出的图像重建模型相应的高效求解算法，算法鲁棒性强、收敛迅速；3) 考虑不同色彩通道间的相关性，将所提出的重建模型扩展至彩色相关图像集重建问题，进一步提升图像重建质量；4) 提出复杂度较低的“预测-重建”结构辅助前述模型完成整个欠采样图像集的重建。本项目的研究，可以为压缩感知在多媒体信息采集领域中的应用提供理论参考和技术支持。经过三年的研究，本项目组发表学术论文13篇，其中SCI检索3篇，EI检索5篇；申请国家发明专利2项，另有3项前期研究成果获授权；获软件著作权登记5项；培养博士研究生2人及硕士研究生6人。