Spectral CT image restoration using average image induced nonlocal means filter

Spectral computed tomography (CT) with energy-resolved detector has gained considerable interests in recent years due to its remarkable abilities in yielding high energy resolution and material differentiation. However, because of the limited photon counts using narrow energy bin, spectral CT images usually suffer from poor SNR and strong streak artifacts. In this work, to restore each energy-specific image, we propose an average image induced nonlocal means (aviNLM) filter by exploiting redundant information in the whole energy domain. Specifically, the average image is generated by averaging all images reconstructed from each energy domain, which is used to induce each energy-specific image restoration. The numerical simulation studies have shown that the present aviNLM filter for spectral CT image restoration can achieve noticeable gains than direct filtered back-projection image reconstruction in terms of noise-induced artifacts suppression.