Bosancica AI
About Project
In collaboration with a National Museum and a National Library, we developed a pilot-stage end-to-end HTR pipeline combining Kraken-based line segmentation and recognition with TrOCR, including geometry based correction of systematic line over-segmentation. We evaluate generalization on an unseen book of manuscripts and conduct an expert pilot study.
The expert pilot found outputs useful as first-pass drafts while identifying visually confusable characters, diacritics, and word-boundary errors requiring expert verification. Machine-generated drafts can therefore support experts in manuscript transliteration.
Mentors: Assist.Prof. Dr. Aida Brankovic, Vahidin Hasić, Kenan Šehić
Collaborators: Farah Piralić, Emir Baručija, Tarik Agić
Partners: https://qla.dev/