基于Contourlet域分块压缩感知的图像融合

Image Fusing by Block Compressed Sensing in Contourlet Domain

针对传统图像融合方法导致纹理细节丢失的现象,提出了一种基于抗混叠移不变Contourlet域的分块压缩感知(block-based compressed sensing,BCS)图像融合算——Contourlet_BCS。把善于表达图像纹理及边缘信息的Contourlet变换引入了压缩感知稀疏表示中,同时对分解得到的低频系数采取加权的区域能量融合规则,高频系数采取基于广义高斯分布模型的加权融合规则进行图像系数融合,最后在压缩感知框架下利用带平滑处理的投影Landweber算法重构。实验结果表明,Contourlet_BCS融合效果优于传统方法,融合的图像纹理清晰.边缘细节信息更为丰富。

For traditional image fusion method results in loss of texture detail, a block compressed sensing image fusion algorithm in contourlet domain with a shift invariant and anti-aliasing ability called Contourlet_BCS was proposed. Contourlet_BCS introduced contourlet transform into the sparse representation step of compressed sensing since its remarkable feature of articulating the texture and edge information, meanwhile, the low-frequency coefficients fused by weighted rule of regional energy and high-frequency coefficients based on the weighted fusion rule by generalized Gaussian distribution model were also used. Finally, the high quality image can be reconstructed by smooth projection Landweber iteration method under the compressed sensing framework. Experimental results show tbat the image fused by Contourlet_BCS was better than the traditional method and the fusion image texture clear and had more abundant edge details.