关键词: 压缩感知,字典学习,关联成像 中图法分类号TN919.8文献标识码A

Abstract: Compressive sensing based differential correlated imaging can reconstruct high quality object image using less number of samples than traditional correlated imaging method,but the former method still has the problem that large computation memory space and it need long image reconstruction time．To solve the problem,this paper proposed an improved differential compressive correlated imaging scheme using the fact that the fluctuations intensity of the optical field influences the contrast quality of the reconstructed image．The scheme can be described as following:First,the measurement data was preprocessed and only the measurement data that is higher than the average value were used to construct the initial dictionary by re-ordering the selected data and extend them to be vector．Then using the training sample,we got the learned initial dictionary D_0as the sensing matrix,next with the learned D0,and using K-means singular value decomposition(K_SVD) algorithm we obtained the updated dictionary D and the corresponding sparse matrix．Finally,the object image information was gotten by orthogonal matching pursuit algorithm．Real experimental data of the object “single slot” and different imaging method were used,the imaging results show that the improved scheme can reconstruct high-definition images using less sample data amount (about 300samplings) than the traditional compressive sensing based differential correlated imaging method (about thousands of samplings)．Thus the new scheme greatly improves the imaging efficiency and image quality,reduces the excessive demands of the system storage hardware,and shortens the image reconstruction time．In addition,another object’s image can be obtained by single optical path measuring provided that the parameters of optical source and distances unchanged,which can reduce the hardware complexity of correlation imaging．All this benefits will make the correlation imaging to be more practical for real applications.

Key words: Compressive sensing,Dictionary construction and learning,Correlated imaging