Handwriting serves as a distinctive behavioral biometric for personal identification, aiding forensic analysts in tackling crimes such as forgery and threats. This research presents a technique for automatic writer recognition from scanned handwriting, utilizing machine learning and pattern recognition. It employs two handcrafted features—Global Binary Model (GBM) and Angle Co-occurrence (ACO)—to evaluate the consistency of handwriting styles, taking into account elements like slant and letter shape. The methodology involves extracting features from binary handwriting images and applying k-Nearest Neighbors (k-NN) with the \(\chi ^2\) distance metric for recognition. A significant benefit of this approach is its reliance on handcrafted descriptors, rendering it appropriate for resource-limited environments. This is in contrast to deep learning-based systems, which necessitate considerable computational resources and labeled data for training. Despite the drawbacks typically associated with handcrafted features, this method incurs very low costs and requires minimal processing time, leading to dependable applications. Testing across six handwriting databases indicates that our approach is computationally efficient, achieving high recognition rates that exceed those of current state-of-the-art methods. For writer identification, we recorded the following rates: \(99.27\%\), \(98.78\%\), \(99.68\%\), \(99.20\%\), \(100.0\%\), and \(100.0\%\). For writer verification, we recorded the following equal error rates: \(0.28\%\), \(0.54\%\), \(1.29\%\), \(2.71\%\), \(1.90\%\), and \(6.71\%\). These results were obtained using writers from the following datasets: Arabic IFN/ENIT (411 writers), English IAM (657 writers), English CVL (310 writers), Dutch Firemaker (250 writers), Chinese CERUG-CN (105 writers), and English/Greek ICDAR2013 (250 writers).
