Static and dynamic classifier fusion for character recognition

The authors introduce a new method for on-line character recognition based on the co-operation of two classifiers, a static one and a dynamic one. In fact, on-line and off-line recognition present very different qualities and small redundancy. Its complementary treatment can bring very interesting results. In their approach, each classifier which operates respectively on static and dynamic character properties, uses the k-nearest-neighbour algorithm. References have been selected previously, using a clustering technic based on dynamic programming, which takes into account the intra-class variability of dynamics characters. This allows data compilation and increases recognition speed. Test data are presented to both classifiers and results are integrated by a static supervisor which provides the final decision. They present the results on their omniscriptor database which count 36 different classes of character and more than 36000 different characters.