Hand Written Character Recognition Using Dynamic Reshaping

In this project, we propose a new approach to hand written character recognition, called dynamic reshaping, which avoids the respective drawbacks of classical neural network approaches. Contrary to the neural network based recognition, our scheme does not need any training set, instead, only very limited character templates are needed. Dynamic reshaping recognition is based on the dynamic transformation of the shape of input character to the shapes of templates stored in the system. Each template characters forms a unique 2-D attracting force vector field, and all the vectors in such field point to the nearest point on the template character. And input characters are treated as elastic bodies. By superimpose an input character on to these vector fields, the input character starts to be stretched by the attracting force, and gradually transforms to the shape of corresponding template character, and we define potential energy of transformed input character as the total elastic force possessed by the character body. Finally, by measuring time spent in transformations, and potential energy of final transformed shape, input characters can be classified into one of the template characters with smallest transforming time and minimum potential energy.