Reconstructing hv-convex images by tabu research approach
Fethi JARRAY
1,2
, Abdessalem DAKHLI
3
and Ghassen TLIG
1
1. Laboratoire CEDRIC- CNAM, Paris, France
Fethi.jarray@cnam.fr , Ghassen_tlik@yahoo.fr
2. Al-Imam Mohamed Bin Saud University, Riyadh, Kingdom of Saudi Arabia
3.Institut Supérieur de Gestion Gabes, Tunisie
Dakhli-ab@voila.fr
Keywords:
Discrete tomography, Images reconstruction, Tabu research.
Abstract
Consider a binary matrix, we call the horizontal projection of row i, the number of ones on
this row. Similarly, we call the vertical projection of column j, the number of ones on column
j. The standard problem of reconstructing a binary matrix from its orthogonal projections is
defined as follows: given the horizontal projection of each row and the vertical projection of
each column find a binary matrix such that the ones fit with the prescribed projections (see
Figure 1). This problem is polynomial and a polynomial time algorithm is proposed [5].
Figure 1. Reconstruction of a binary matrix from its projections
In general, the standard problem is underdetermined and many solutions may exist. To
reduce the space of feasible solutions, prior knowledge about the image to reconstruct should
be considered. Much real knowledge are considered in literature such as periodicity,
convexity, connectedness, etc. In this work we deal with the reconstruction of hv-convex
binary matrices from their orthogonal projections. A matrix is called hv-convex if the ones
occur consecutively in a single block in each row and in each column. Barcucci et al. show
that this problem is NP-complete [1]. May approximate method has been propose to
reconstruct nearly hv-convex matrices [1, 2, 3]. Dahl and Flatberg [2] provide approximate
solution based on a lagrangean decomposition. Jarray et al. [3] provide an approximate
algorithm based on longest path and network flow algorithm. Recently Jarray and Tlig [4]
use simulated annealing algorithm to reconstruct hv-convex images.
1