Moments and invariant moments are important features used in identification and inspection of industrial parts. It is necessary to compute geometric moment's values in real-time rate. The efficient computation of two-dimensional geometric moments on gray-level images is addressed in this paper. Despite the existence of many algorithms of fast computation of moments, it cannot be implemented for real-time computation to be run on a PC without the use of some special dedicated hardware tools. The reason beyond this is that those fast algorithms do reduce the complexity of computing but still one needs to use floating-point arithmetic operations in the computation process. To achieve real-time computation on a PC machine, what the algorithm suggested here is based on dividing the image into equally sized blocks. This algorithm works by computing local moments at each block using integer operations, then accumulating the total image moments with floating-point operations. With this computation scheme no approximation is used, it is an exact computation. Overcoming this overflow problem, however, is not straightforward without using some kind of transformation to each block. Hatamian's (improved) filter is used to compute those block moments (BLMs) efficiently. The experiments show that the algorithm presented in the paper has greatly reduced floating-point operations in fast computation of moments, and greatly improved the speed of the computation of moments. The new algorithm can be effectively used in real-time identification and inspection of complicated industrial parts.