Turkish handwriting recognition with deep learning methods

Handwritten Character Recognition uses algorithms that can detect characters as images and analyze these images to process them in text format. This technology plays a significant role in various fields, such as digitizing and archiving documents. In the literature, studies on handwritten character recognition generally focus on the English alphabet and commonly used Latin characters. The limited availability of datasets developed for languages with special characters, such as Turkish, can negatively affect the performance of recognition systems in these languages. The aim of this study is to contribute to the literature in the field of handwritten character recognition by preparing datasets containing special characters from the Turkish alphabet and to evaluate the performance of deep learning techniques on our own dataset. For this purpose, a dataset was created by extracting handwritten characters from documents voluntarily filled out by 400 employees in an institution, each with different writing styles. Then, deep learning-based CNN (Convolutional Neural Network) models were trained with this dataset. When the test results of the CNN models were compared, it was found that the most successful model achieved a test accuracy of 93.86%. The performance of the best CNN model was evaluated using various metrics, and the classification success of the model was analyzed in detail.