Although geometric moments have been successfully used as image features in many applications, and hence widely studied in the literature, their computation, is too expensive, which limits their use in real-time tasks such as visual servoing. This paper describes a new formulation for computing the geometric moments of a 2-D gray-level image. It is simpler than Li's approach, which not only leads to the most efficient algorithm for moment computation known until now, but which is also usable in real-time applications due to the ease of its parallel implementation. Moreover, a method to solve the problem of image reconstruction from a finite set of its geometric moments, that outperforms previous approaches, is directly derived from this formulation. The computational costs of the different operations are also evaluated, showing that they efficiently reduce both addition and multiplication complexity of direct geometric moment computation.