Handwritten Digit Recognition using Fine-Tuned Convolutional Neural Network Model

Traditional handwriting recognition algorithms were reliant heavily on existing knowledge and personalized features. It is more difficult to train a computerized optical character recognition (OCR) system using these criteria. Recent advances in the identification of handwriting research have been made possible by using deep learning techniques as their main mechanism. But when the volume of written content quickly increased and incredibly powerful computing resources became available, digit recognition accuracy had to have deteriorated, mandating more applicants’ implementation in order to yield accurate and precise results. To develop an effective approach for resolving handwriting detection challenges, this work uses the convolutional neural network (CNN) model which demonstrated the outperforming results in interpreting the structure of handwritten characters as well as phrases. This model is also capable of reducing the operational complexity and cost by implementing this model on different optimizers namely, Adam and SGD. The results have been evaluated at different numbers of epochs which identified that the proposed model outperforms with Adam optimizer by showing the highest accuracy of 98% at epoch 50. This model can be further implemented by varying the learning rate values to achieve improved accuracy and reducing the loss.