An object-based tomographic recostruction algorithm exploiting interest points in image projections

In this paper we propose an algorithm for the reconstruction of binary images composed by geometrical objects, specifically formed by circles inside a ring. We studied how features of these objects are reflected onto the projections of the image itself, and how by an accurate analysis of the projections, even from a single direction, we can obtain a good amount of information about the position of the circles by searching for interest points. In this case, by computing the discrete derivatives of the projections we can often determine with precision the positions of the circles in the original image. We compared the experimental results of our procedure with previous works to prove how our algorithm obtains very promising results. Finally, we discuss how a similar approach could be extended to many other more complex classes of images that may describe real world applications.