CNN-based Thai handwritten OCR: an application for automated mail sorting

Automated mail sorting machines commonly use Optical Character Recognition (OCR) to recognize destination postcodes on envelopes and sort them accordingly. However, errors in postcode transcription can lead to misdirected postal items. To address this issue, we propose a system that sorts Thai handwritten envelopes by recognizing the postal district name within the destination address block instead of relying solely on postcodes. Our approach relies on computer vision solutions that extract handwritten words from the envelopes. AlexNet and ZFNet are used for OCR-based feature extraction. Through experimental results, we observe that the ZFNet model slightly outperforms AlexNet in classifying district names. While our proposed system shows promising results, further refinement is necessary for optimal performance.