Hybrid Intelligence: Filtering and Deep Learning for Handwritten Text Recognition

Handwritten text recognition systems have gained substantial attention in pattern recognition and artificial intelligence due to their applications in digitizing historical documents and automated reading tasks.This study proposes a novel preprocessing approach for recognizing Kufic manuscripts, aiming to improve the accuracy of handwritten text segmentation.The approach consists of three stages: noise removal, thresholding, and additional noise removal after thresholding.Hybrid filtering, using Wiener and Median filters, effectively reduces noise in the input images.The u-Net deep learning model is employed for precise thresholding and segmentation of the handwritten text.The post-thresholding noise removal step refines the segmented regions and eliminates residual noise artifacts.The proposed approach contributes to enhancing the Quranic Kufic Manuscripts Recognition System (QKMRS) and advancing the digitization of historical documents and the preservation of cultural heritage.The performance measures for noise removal include Mean Square Error (MSE) and Peak Signal Noise Ratio (PSNR).For segmentation and thresholding Intersection over union (IoU), entropy and dice coefficient are used.