Segmentation in an ancient document imaging and characters

Аннотация

Transcribing Ancient Manuscript Detection (TAMD) is a critical task for enabling the use of manuscript images for retrieving information in an Old Uzbek multiple image manuscripts. Precise detection of specific artefacts such as pixel saturations, motion blur, specular reflections, bubbles and instruments is essential for high-quality frame restoration. This work describes our model on our limited TAMD dataset to detect Arabic texts in respect to images. Our method is based on YOLOv5 for detection and UNet++ for segmentation and applied some modifications to Arabic. Our method reaches a mAP of 0.290 for the detection task over all classes of artefact for 203 images. Segmentation model achieves. 0.285 IoU value..