The goal of this paper is the reconstruction of topo-logically accurate 3-dimensional triangular meshes representing a complex, multi-layered plant tissue structure. Based on time sequences of meristem images of the model plant Arabidopsis thaliana, displaying fluorescence markers on either cell membranes or cell nuclei under confocal laser scanning microscopy, we aim at obtaining faithful reconstructions of all the cell walls in the tissue. In the presented method, the problem is tackled under the angle of topology, and the shape of the cells is seen as the dual geometry of a 3-d simplicial complex accounting for their adjacency relationships. We present a method for optimizing such complexes using an energy minimization process, designed to make them fit to the actual adjacencies in the tissue. The resulting dual meshes constitute a light discrete representation of the cell surfaces that enables fast visualization, and quantitative analysis, and allows in silico physical and mechanical simulations on real-world data.