کنفرانس مهندسی برق ایران

صفحه اصلی / سی و دومین کنفرانس بین المللی مهندسی برق

Spotting of a Particular Printed Word in Farsi Handwritten Forms Using Deep Learning

نویسندگان :

Mohammad jafar Gholami Kenari¹ Ehsanollah Kabir²

1- دانشگاه تربیت مدرس تهران 2- دانشگاه تربیت مدرس تهران

کلمات کلیدی :

Word Spotting،Mask R-CNN،Anchors،Data Augmentation،Farsi،Persian

چکیده :

Keyword spotting in documents plays a crucial role in information retrieval and document analysis. Recent years have witnessed significant progress in keyword spotting through deep learning methods. This paper introduces a method that utilizes a pre-trained Mask R-CNN with transfer learning to spot the printed keyword “تاریخ” in the printed forms filled in handwriting. To address data scarcity and enhance the network's performance, data augmentation methods are employed. Additionally, specific to the keyword “تاریخ”, adjustments such as changes in the dimensions and aspect ratio of anchors, are implemented in the region proposal network (RPN). The results illustrate that the proposed method achieves a mean Average Precision (mAP) of 98.1 percent