Robust reconstruction for fluorescence molecular tomography based on correntropy matching pursuit

Fluorescence molecular tomography (FMT), as a promising imaging modality in preclinical research, can obtain the three-dimensional location information of the specific tumor for small animal imaging. A lot of methods based on orthogonal matching pursuit (OMP) have been used for FMT reconstruction. OMP and most of its variants use the mean square error criterion to estimate the sparse vector which depends on the Gaussian assumption of the error distribution. However, this method has poor performance on reconstruction with non-Gaussian noise. In this study, we propose correntropy matching pursuit (CMP) method to alleviate this problem of OMP. Unlike some other matching pursuit methods, CMP method is independent of the error distribution. This method can assign small weights on severely corrupted entries of data and large weights on clean ones adaptively, thus reducing the effect of large noise. The Gaussian assumption of the noise distribution is often used in the matching pursuit (MP) methods so that they can't work well with the non- Gaussian noise. Our method is independent of the noise distribution and has better performance on FMT reconstruction. Compared with StOMP and SASP, CMP method shows better performance on FMT reconstruction with non-Gaussian noise and similar performance on FMT.