Image Processing on Projections and Its Applications for Image Reconstruction

Reconstruction tomography produces 2-dimensional cross-sections of 3-dimensional objects from their projections taken from several directions. Depending on the reconstruction task, projections can be acquisited by X-rays, gamma-rays, etc. In the most common situation radiation is generated outside the object (CT), in other scenarios the object itself emits gamma-rays (SPECT, PET). In image reconstruction the task is – roughly speaking – to backproject the sinogram consisting of the projections of the original image. This is usually performed by filtered backprojection (FBP) or algebraic iterative methods (ART, SART, SIRT). In practice, projections are taken from discrete angles, and they are noisy, too. The aim of this project is to improve the well-known methods of continuous and discrete tomography, and especially the followings:

• study of image processing methods, that can be applied directly on the projections to improve reconstruction quality (active contours, segmentation, edge-detection, denoising, etc.)
• incremental reconstruction by identifying important projection angles
• improving techniques to measure the quality of the reconstructed images (similarity to model image, statistical features, smoothness, etc.)
• measuring the information content of projections
• determination of attenuation coefficients from the sinogram for certain classes of images
• developing efficient discrete image reconstruction methods of novel types, based on parallel processing using GPUs where possible