Efficient computation of moments for pattern recognition

Two algorithms for the realization of two-dimensional image moments are proposed. The first algorithm groups the realization into row-oriented moments and column-oriented moments and then uses simple additions for the major part of the computation. The second algorithm divides pixels into blocks and uses a simple table look-up technique for its realization. The basic formulation of the second algorithm is then further extended to a full row-oriented realization. This last formulation simplifies the whole realization procedure and gives the best timing for practical realization. These approaches give 20- to 40-fold reduction in the numbers of operations compared to other algorithms, and also have much better performance in practical realizations