A Novel Scheme for Implementation of the Scanning ntuple Classifier in a Constrained Environment

The scanning ntuple classifier is an efficient and accurate classifier for handwriting recognition. One of the major difficulties in implementing this scheme is its demand for a very large memory space, thus making it unsuitable for resource constrained systems such as embedded applications. This paper proposes some modifications to the basic sntuple algorithm which eliminates the necessity of normalizing the chain-code length, by adjusting the memory cell increments as an inverse function the chain length. The resulting system performance is shown to be superior to the standard sntuple configuration in both speed and accuracy when smaller and fewer sntuples are used, a configuration which also reduces the demand for memory.