A Deep Learning Approach for Evaluating Children's Handwriting

Facilitating precise writing proficiency in early learners is crucial for their foundational development and cognitive abilities. While traditional handwriting instruction methods are valuable, they may not always provide the specific validation needed to ensure accuracy. Traditional handwriting methods can be slow and often lack personalized feedback, which is important for effective learning. This study presents a method for teaching young learners how to write correctly by offering an interactive user interface in which new learners try to imitate the displayed word while the model checks their imitation and provides constant feedback on the handwritten word at the character level. Assessing children's handwriting is tricky compared to adults because their letters can be hard to distinguish. Creating a model to evaluate children's handwriting could change how we teach and evaluate writing skills. It could make education more efficient and effective. This paper explores various methods for evaluating handwritten words at the character level. First, the K-nearest neighbors (KNN) model is considered, known for its ability to distinguish similar and dissimilar items. However, KNN faces challenges in handling numerous classes and adapting to new data. The second approach is considered to have a Siamese-based twin network consisting of three CNNs and two LSTM layers. This model exhibits promise, achieving over 90% accuracy on training and validation sets. Nevertheless, it encounters difficulties with new, unseen data. The approach considered in this paper presents a simple yet robust model, featuring three CNN layers, one flattened layer, and two dense layers. This model effectively classifies all 52 character classes. Notably, the CNN-based model attains an impressive 96% accuracy on both test and validation sets, even when dealing with handwriting style it has not encountered before. This demonstrates the model's ability to apply learned knowledge and potential for practical use.