A Review of Handwritten and Printed Text Segmentation: Problem Formulation, Benchmarks, Methods, and Open Challenges

Document image analysis transforms digitized documents into machine-readable representations. Within this field, Handwritten and Printed Text Segmentation (HPTS) separates handwritten strokes from printed text in mixed document images and supports downstream tasks such as optical character recognition, form understanding, and historical document restoration. This paper presents a structured narrative review of representative HPTS studies from four perspectives: problem formulation, methodological evolution, data resources and evaluation, and open challenges. We first clarify the technical scope, output representations, and objectives of HPTS, and then trace representative methods from multi-stage pipelines based on handcrafted features to deep learning-based methods. Beyond tracing methodological progress, we analyze the empirical foundations of HPTS, focusing on how dataset construction, annotation design, benchmark coverage, and evaluation protocols affect the validity and cross-study comparability of reported results. We characterize representative datasets in terms of document domain, construction strategy, annotation granularity, label ontology, and validation scope. We then assess benchmark reuse across representative studies and compare published results only under compatible experimental settings. The analysis reveals that datasets with dense annotations for handwritten–printed overlap remain scarce and that private or study-specific data are still widely used. Differences in label spaces, data splits, prediction targets, and evaluation metrics therefore prevent a reliable field-wide ranking of existing methods. Based on these observations, we highlight several open challenges for future HPTS research, including limited benchmark diversity, inconsistent annotation and evaluation protocols, insufficient cross-domain generalization, and limited integration with downstream document-intelligence tasks.