Abstract:
Handwritten Digit Recognition (HDR) is a widely used technology demonstrated
in various research topics. However, there are many limitations because everyone's
handwriting differs, including important handwriting characteristics. The experimental
results of the previously proposed methods were unsatisfactory when using limited
amounts of the datasets for the training model and inefficient pre-processing.
Therefore, we describe a new method for increasing the recognition rate through well-planned pre-processing and enhanced feature extraction for applying Particle Swarm Optimization (PSO). Compared to other techniques, the proposed method performed
well when tested against individual handwriting (INDV Datasets), the original MNIST
database, and Handtyped datasets. Our results are a high recognition rate in testing,
including short runtime, and within the scope of the proposed work that has been set to solve the limitations that arise with the method. Current work and increase the efficiency of the work presented even further.