If an unknown finite set C ⊂ Z 2 is cut by lines parallel to given directions, then one may count the number of points of C that are intercepted by each line, that is, the projections of C in the given directions. The inverse problem consists in reconstructing the set C , interpreted as a binary image, from the knowledge of its projections. In general, this challenging combinatorial problem, also related to the tomographic reconstruction of an unknown homogeneous object by means of X-rays, is ill-posed, meaning that different binary images exist that match the available projections. Therefore, as a preliminary step, one can try to find conditions to be imposed on the considered directions in order to limit the number of allowed solutions. In this paper we address the above problems for sets C contained in a finite assigned lattice grid, and generalize some results known in literature. First, we describe special sets of lattice directions, called simple cycles, and focus on some of their properties. Then we prove that uniqueness of reconstruction for binary images is guaranteed if and only if the line sums are computed along suitable simple cycles having even cardinality. As a second item, we prove that the unique binary solution can be explicitly reconstructed from a real-valued solution having minimal Euclidean norm. This leads to an explicit reconstruction algorithm, tested on four different phantoms and compared with previous results, which points out a significant improvement of the corresponding performance.

In discrete tomographic image reconstruction, projections are taken along a finite set S of valid directions for a working grid 𝒜. In general, uniqueness cannot be achieved in the whole grid 𝒜. Usually, some information on the object to be reconstructed is introduced, that, sometimes, allows possib le ambiguities to be removed. From a different perspective, one aims in finding subregions of 𝒜 where uniqueness can be guaranteed, and obtained in linear time, only from the knowledge of S. When S consists of two lattice directions, the shape of any such region of uniqueness, say ROU, have been completely characterized in previous works by means of a double Euclidean division algorithm called DEDA. Results have been later extended to special triples of directions, under a suitable assumption on their entries. In this paper we remove the previous assumption, so providing a complete characterization of the shape of the ROU for such kind of triples. We also show that the employed strategy can be even applied to more general sets of three directions, where the corresponding ROU can be characterized as well. Independently of the combinatorial interest of the problem, the result can be exploited to define in advance, namely before using any kind of radiation, suitable sets of directions that allow regions of interest to be included in the corresponding ROU. Results have been proved in all details, and several experiments are considered, in order to support the theoretical steps and to clarify possible applications.

In this paper the well-known problem of reconstructing hv-convex polyominoes is considered from a set of noisy data. Differently from the usual approach of Binary Tomography, this leads to a probabilistic evaluation in the reconstruction algorithm, w