Answers are grounded in this paper and cite their sources.
Background. MODI is a cursive historical script used extensively for Marathi writing, and its handwritten documents present challenges for optical character recognition because of cursive connections, visual similarity, writer variation and document degradation. Objective. This review consolidates evidence on handwritten MODI recognition with specific attention to barakhadi, defined here primarily as consonant–vowel combinations, while distinguishing such combinations from datasets that contain only individual consonants and vowels. Information sources. Two supplied reference compilations were combined with exploratory searches of open web and repository sources, including IEEE DataPort, Mendeley Data, institutional repositories, arXiv and publisher pages. Searches were last run on 24 October 2026. Eligibility criteria. Studies and dataset records concerning handwritten or historical MODI recognition, segmentation, transliteration, datasets, or barakhadi/consonant–vowel recognition in related Indic scripts were eligible. Printed-only OCR and duplicate or insufficiently identifiable records were excluded. Synthesis methods. A narrative thematic synthesis was performed. Dataset characteristics, recognition level, methods, preprocessing and reported results were tabulated where information was available. Reporting completeness was examined for seven representative studies. No pooled statistical estimate or formal risk-of-bias assessment was performed. Results. Forty-three records were retained: 36 papers, 3 dataset records, 2 theses and 2 preprints. Direct MODI barakhadi studies identified in this review are concentrated in one research group. Reported recognition results are promising, including 94.552% test accuracy for a 7,721-image MODI barakhadi dataset and 95.7% top-1 accuracy for a 360-class MODI character study; however, these figures are not directly comparable because datasets, class inventories, writer populations, preprocessing and evaluation protocols differ. Limitations. The review is narrative rather than a registered systematic review. Screening was performed by one reviewer with AI assistance, excluded search hits were not logged, the search was exploratory rather than exhaustive, and several records were available only as abstracts or repository records. Conclusions. The literature lacks a standardised, large and writer-independent MODI barakhadi benchmark with a documented class inventory and reproducible evaluation. Future work should compare flat and decomposed barakhadi recognition on common data and connect isolated-unit recognition with word-, line- and document-level OCR.
No takes yet. Share an insight, caveat, or question.
Gadpade et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: