AI-based methods and hyperspectral imaging techniques towards an automated translation of ancient Egyptian hieroglyphs: perspectives and open problems

In recent years, an interdisciplinary study branch, connecting Egyptology with the technological domains of artificial intelligence (AI) and advanced imaging techniques has been started, with the ultimate goal of providing new tools for automating translation of ancient Egyptian texts. Indeed, the archeological documentary heritage on ancient Egypt comprises a huge amount of still unpublished texts and documents, whose extensive translation remains highly challenging and time-consuming. It is therefore evident that providing archeologists with new methods for automating - or at least for semi automating – massive translation of ancient Egyptian hieroglyphs would represent a breakthrough technological achievement and an invaluable support to the Egyptological research. The recent explosion of AI-technologies across all the fields of knowledge has boosted the exploration of AI-based approaches to tackle the multifaceted problem of Egyptian hieroglyphs decoding. In earlier works, the authors proposed re-trained architectures of well-established Convolutional Neural Networks (CNNs) to address the segmentation and classification of Egyptian hieroglyphs. Even though the preliminary results were encouraging, a lack of representative training data-sets of symbols emerged as a primary roadblock. Other challenges to solve are related to the high variability among writing styles (e.g. different scribe's hands), different symbols systems (e.g. hieroglyphs, hieratic, demotic), materials diversity and different supports (papyri, stone, or wood) and degraded surfaces compromise the readability of characters. All these issues make the automated recognition task even more complex in the case of images acquired from real archaeological items. Then, as a subsequent step in this research, use of hyperspectral imaging (HSI) was proposed to obtain augmented images of degraded inscriptions on archeological surfaces. The enhancing capability of HSI was successfully exploited to improve the performance of selected retrained CNN in addressing the tasks related to the automated hieroglyphs recognition. In the present contribution the research recently carried out by authors will be reviewed, by presenting the conceptual path hitherto followed and the on-going studies: the main focus of the discussion will be the combined use of re-trained CNNs architectures with HSI techniques, looking at the perspectives and the open problems.