Dictionary-based super resolution for positron emission tomography images
In this paper, a strategy to increase the resolution of positron emission tomography (PET) images, using a previously trained high resolution dictionary for the sinograms is proposed. The low resolution input sinogram is divided into patches of 5x5 samples. The sparse code of each patch is calculated and applied to the high resolution dictionary to obtain the best high resolution patch. The estimated high resolution sinogram is processed by the filtered backprojection (FBP) or by the ordered subsets expectation maximization (OSEM) reconstruction algorithm. Results show that, in both cases, the dictionary method outperforms the bicubic interpolation method by more 3% in PSNR. OSEM algorithm yields even better results than the FBP algorithm. However, the reconstruction time is exacerbated.