Arabic Handwritten Passage Length Classification Using Geometric Features and Linear and RBF-SVM Models

Arabic handwritten text classification remains challenging due to variations in writing styles, connected characters, word spacing, and image dimensions. This study investigates the effectiveness of Radial Basis Function Support Vector Machine (RBF-SVM) and Linear Support Vector Machine (Linear SVM) algorithms for classifying Arabic handwritten text images according to passage length. The experimental evaluation was conducted using a dataset comprising 11,375 handwritten Arabic text images and their corresponding transcriptions, collected from 836 different writers. Based on transcription length, the samples were divided into three classification categories: short passages (1–10 words), medium passages (11–15 words), and long passages (16 or more words). Four geometric features were extracted to characterize the spatial properties of the handwritten samples: image width, image height, aspect ratio, and logarithmic pixel area. A writer-independent data partitioning strategy was employed to generate training and testing sets containing 9,096 and 2,279 images, respectively. Classification performance was evaluated using accuracy, balanced accuracy, precision, recall, F1-score, and confusion matrix analysis. The results showed that RBF-SVM achieved an accuracy of 63.05% and a balanced accuracy of 44.42%, slightly outperforming Linear SVM, which achieved corresponding values of 62.44% and 43.29%. RBF-SVM correctly classified 1,437 test images, compared with 1,423 for Linear SVM. Both classifiers demonstrated their strongest classification performance for medium-length passages, whereas long passages were more difficult to distinguish. Overall, the findings indicate that replacing a linear kernel with a radial basis function kernel provides only marginal performance improvements. Furthermore, the relatively low balanced accuracy suggests that geometric features alone are insufficient to reliably distinguish between passage-length categories. This study establishes a baseline for Arabic handwritten text classification based on passage length and highlights the need to incorporate more discriminative visual and structural features to improve classification performance.