The authors combine statistical and structural techniques forautomatic segmentation and recognition of zip codes. This kind ofcombination has been advocated by several authors. The aim is toincorporate advantages of both approaches, that is to say, theflexibility and the adaptability to noisy patterns, for statisticaltechniques, and possibility to describe very complex and specificpatterns, using syntactic techniques. The statistical stage is the firstand main stage of the authors recognition system. It uses concavitymeasurements to characterize the characters and a linear discriminationtechnique is implemented to affect one character to one cluster. Thesecond stage, called structural supervisor, is developed in order tocontrol and improve the results of the first stage
展开▼