Maximum a posteriori tomographic image reconstruction with a dense image patch prior
演題番号 : P1-r15
Atsunori Kanemura:1 Masa-aki Sato:1
1:ATR Neural Information Analysis Laboratories
The teqhnique of tomographic image reconstruction, or computed tomography (CT), is capable of generating tomographic images without cutting the object into slices. Despite of its history of research since the advent of x-ray CT in the mid-1960s, the technique has not completed and is still making progress to be a more reliable and safety imaging modality.We propose a maximum a posterior estimation method for CT reconstruction by using a prior distribution that is defined on dense overlapping patches of tomographic image. The prior has local adaptivity and thus it can preserve fine structure as well as smoothing non-textured regions. The parameters for the dense-patch prior are learned from given observations.The performance of the proposed CT algorithm is tested on several simulation experiments under fewer projections and moderate noise, where classical CT algorithms like filtered back-projection fail.The proposed method has potential applicability to SPECT, PET, and MRI because they share the problem structure of reconstructing tomograms from projections.