The recognition of Roman hand-written characters using dynamic information recovery

The use of temporal information such position, velocity and acceleration has led on-line systems to achieve better performance than off-line systems. Some research has been undertaken in order to transform the off-line problem into an on-line one. The paper shows that the Chinese postman's paradigm can successfully be used to extract consistent sub-stroke ordering in Roman writing. An algorithm called optimal trace recovering (OTR) is presented which incorporates the Chinese postman's paradigm and is shown to produce an optimal writing trace with respect to minimum stroke length and deviation at special (cross and fork) points. The OTR algorithm produces an acceptable consistency in recovering the writing trace for each writer and furthermore the method described is computational fast and robust. Dynamic features has been extracted from the directed graph produced by the OTR algorithm and used by a statistical classifier. On average, recognition rates of 91.64% (first choice) and 94.55% (first two choices) has been achieved.