谱间和帧内协同稀疏的高光谱图像压缩感知模型

发布时间：2019-01-06 16:53

【摘要】：本文基于压缩感知理论,利用高光谱图像谱间存在极强的相关性,提出一种基于谱间和帧内协同稀疏的高光谱图像压缩感知模型(hyperspectral image collaborative sparsity measure,HICoSM).模型包括对高光谱图像各波段帧内及其帧间3方面的稀疏性挖掘:一是对各波段帧基于局部区域平滑性的稀疏性度量;二是对各波段帧基于非局部区域纹理、边缘等细节信息的自相似性稀疏性度量;三是相邻波段帧基于谱间相关性的预测稀疏性度量,具体利用前一个波段帧,通过最小二乘法线性预测形成当前波段帧的预测帧,通过确定预测帧与当前波段帧的最佳预测差实现谱间的稀疏性度量.进一步,给出了所提出模型的数值计算过程.仿真实验表明,模型HICoSM在对各个波段帧的稀疏性进行度量的基础上,通过挖掘和测量高光谱图像波段间的谱间稀疏性,有效地提高了各波段帧在压缩感知恢复阶段的解码质量.
[Abstract]:Based on the theory of compression sensing, a hyperspectral image compression sensing model (hyperspectral image collaborative sparsity measure,HICoSM) is proposed, which is based on the strong correlation between the spectra of hyperspectral images. The model includes three aspects of sparse mining in and between frames of each band of hyperspectral images: first, the sparse measurement of each band frame based on local region smoothness; The second is the self-similarity sparsity measure of each band frame based on the detail information of non-local region texture and edge. The third is the prediction sparsity measure of adjacent band frames based on the correlation between spectra. Using the previous band frame, the prediction frame of the current band frame is formed by the least square method. The sparsity measure is realized by determining the optimal prediction difference between the predicted frame and the current band frame. Furthermore, the numerical calculation process of the proposed model is given. The simulation results show that the model HICoSM can effectively improve the decoding quality of each band frame at the stage of compressed perceptual recovery by mining and measuring the spectral sparsity between bands on the basis of measuring the sparsity of each band frame.
【作者单位】：辽宁师范大学计算机与信息技术学院;辽宁师范大学自然地理与空间信息科学辽宁重点实验室;
【基金】：国家自然科学基金项目(批准号:41271422,61402214) 高等学校博士学科点专项科研基金项目(批准号:20132136110002) 辽宁省博士科研启动基金项目(批准号:20121076) 辽宁省教育厅科学研究一般项目(批准号:L2013405,L2013406)资助
【分类号】：TP751

【相似文献】