Data-Driven Problem Reduction for Image Reconstruction From Projections Using Gift Wrapping

A generic data-driven approach called focus of attention is presented that automatically segments first the projection data and then the image into object and background prior to reconstruction. This segmentation, which is based on the formation of convex hulls, subsequently allows the computational resources to be devoted to the object data while the background data is discarded from further consideration. Experimental results are presented for high-resolution reconstruction of a cone beam x-ray CT image of a laboratory mouse.