
Handwritten Text Segmentation &
Optical Character Recognition
with Visual Transformer
I focused on the part of OCR called ‘Handwriting Line Segmentation’ to detect and segment text lines pixel-wise one by one from a given handwriting paper image. I used the Convolutional Neural Network model to train with handwritten images and their corresponding line masks to detect lines at a later step. As a final decision, I end up using U-net to segment every single pixel of the images. I used a mask that has two classes one background and one handwriting line where the background is black, and the handwriting line is white.