Discrete tomography focuses on image representation by its discrete projections, and the related inversion algorithms (or image reconstruction). Our study is based on redundant representations (considering more than just few projections). We propose a new approach to compute further redundancy (i.e. new projections) from a set of existing projections. While this technique relies on the geometric properties of ghosts, which are elements of the 2D image that sum to zero following some projection directions, we show an equivalent method using 1D convolutions, thus avoiding the explicit image reconstruction. This technique has interesting applications in distributed storage systems, where the use of redundancy data is key for system reliability.