Similar handwritten Chinese character recognition based on adaptive discriminative locality alignment

Discriminative locality alignment (DLA) has been successfully applied in similar handwritten Chinese character recognition (SHCCR). But, the performance of DLA heavily depends on the choice of parameters and the optimal parameters among different groups of similar characters are not consistent. To address this problem, we present an improved method with few parameters, called adaptive discriminative locality alignment (ADLA), whose optimal parameters are the same for different groups of similar characters. Further, the kernel discriminative locality alignment (KADLA) is formulated. The experimental results demonstrate that ADLA has higher performance than DLA in recognition rate, and KADLA has even higher recognition rate. In practice, Since KADLA involves much more time and storage cost, ADLA is a better choice for SHCCR.