In the discrete topology optimization, material state is either solid or void and there is no topology uncertainty caused by any intermediate material state. In this paper, the improved hybrid discretization model is introduced for the discrete topology optimization of structures. The design domain is discretized into quadrilateral design cells and each quadrilateral design cell is further subdivided into triangular analysis cells. The dangling and redundant solid design cells are completely eliminated from topology solutions in the improved hybrid discretization model to avoid sharp protrusions. The local stress constraint is directly imposed on each triangular analysis cell to make the designed structure safe. The binary bit-array genetic algorithm is used to search for the optimal topology to circumvent the geometrical bias against the vertical design cells. The presented discrete topology optimization procedure is illustrated by two topology optimization examples of structures.