Accurate Handwritten Arabic Textual Amount Recognition via a Robust Fuzzy Nearest Neighbor Framework: Model Development and Validation

Background: Recognizing handwritten Arabic textual amounts presents persistent challenges due to the cursive nature, varied writing styles, and contextual dependencies inherent in the Arabic script.Addressing this task is particularly critical in automated financial document processing, where high accuracy is paramount.Methods: This paper introduces a Fuzzy Nearest Neighbor (FNN) framework tailored for recognizing handwritten Arabic amounts.The approach integrates an adaptive preprocessing pipeline-comprising binarization, noise reduction, and segmentationwith a hybrid feature extraction method that combines structural and statistical descriptors.These features are then projected into a fuzzy space and classified using a fuzzy logic-enhanced nearest neighbor strategy.Results: Using a curated dataset of handwritten Arabic check amounts, the proposed model achieved a recognition accuracy of 97.4%, surpassing the performance of traditional classifiers such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and K-Nearest Neighbors (k-NN).The system demonstrated robust generalization capabilities, even when presented with degraded and previously unseen inputs.Conclusions: The proposed FNN model offers an interpretable and scalable solution for handwritten Arabic text recognition.Its robustness to stylistic variations and noise makes it well-suited for real-world applications in banking, postal automation, and historical document processing.