Recognition of Egyptian hieroglyphic texts through focused generic segmentation and cross-validation voting

Ancient Egyptian hieroglyphs form part of a complex language that has attracted the attention of Egyptologists, historians, and amateurs for centuries. In use for more than 3000 years, they consist of hundreds of symbols that can be transcribed into their Latin phonemes. Although there have been som...

Descripción completa

Detalles Bibliográficos
Autores: Fuentes Ferrer, Raúl, Duque Domingo, Jaime, Herrera Caro, Pedro Javier
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universidad Nacional de Educación a Distancia
Repositorio:e-spacio. Repositorio Institucional de la UNED
Idioma:inglés
OAI Identifier:oai:e-spacio.uned.es:20.500.14468/26492
Acceso en línea:https://hdl.handle.net/20.500.14468/26492
Access Level:acceso embargado
Palabra clave:1203.17 Informática
egyptian hieroglyphs
generic segmentation
SAM
CVV
ConvNeXt
Descripción
Sumario:Ancient Egyptian hieroglyphs form part of a complex language that has attracted the attention of Egyptologists, historians, and amateurs for centuries. In use for more than 3000 years, they consist of hundreds of symbols that can be transcribed into their Latin phonemes. Although there have been some previous works on the recognition of hieroglyphs through computer vision, this is a study of unprecedented depths and presents several unique contributions. On the one hand, we have created the largest and most complete dataset of existing Egyptian hieroglyphs to date, covering all the main symbols used on stelae. On the other, we have carried out a systematic analysis of detection, segmentation, and classification methods, focusing our research on a composite method of focused generic segmentation and classification with an ensemble model of ConvNeXt backbones using Cross-Validation Voting (CVV). Our trained model has been evaluated against several carved or painted stone stelae, obtaining excellent results. To the best of our knowledge, there is currently no other methodology capable of obtaining the classification results presented in this paper, and the method and the dataset presented represent a very significant advancement in the development of automated methods for reading Egyptian hieroglyphic texts.