Handwriting Detection Using Object Detection Models

In this paper, we propose a new method for handwriting text detection using the YOLOV5 object detection neural network. Our approach accurately identifies text sequences regardless of orientation, size, or type (handwritten or printed) and locates handwritten text regions within images. We then utilize the TrOCR model by Hugging Face, which employs an encoder-decoder mechanism to extract and digitize text from the images. To train and evaluate our model, we introduce a clinical receipts dataset tailored for handwriting text detection, addressing challenges like variations in styles, sizes, and orientations. Through extensive experimentation, we demonstrate our approach's effectiveness on benchmark datasets, achieving competitive performance against existing methods. Additionally, we explore potential applications of handwriting text detection in tasks such as OCR, document classification, and information extraction, presenting a promising solution for robust and efficient handwriting text detection in diverse real-world scenarios.