An empirical study of reconstructing hv-convex binary matrices from horizontal and vertical projections

The reconstruction of hv-convex binary matrices (or equivalently, binary images) from their horizontal and vertical projections is proved to be NP-hard. In this paper we take a closer look at the difficulty of the problem. We investigate different heuristic reconstruction algorithms of the class, and compare them from the viewpoint of running-time and reconstruction quality. Using a large set of test images of different sizes and with varying number of components, we show that the reconstruction quality can depend not only on the size of the image, but on the number and location of its components, too. We also reveal that the reconstruction time can also be affected by the number of the so-called switching components present in the image