Exploring Hieroglyph Recognition: A Deep Learning Approach

Hieroglyphs are an ancient pictorial writing system used for conveying messages. This research paper aims to introduce a novel hand-drawn hieroglyph dataset based on the Gardiner classification - a standard reference for studying ancient Egyptian hieroglyphs. The prepared dataset comprises 59,514 images featuring 763 distinct symbols. To demonstrate the robustness of the data, a comparative analysis was performed on various convolutional neural network architectures trained on the dataset. Including popular models like ResNet, VGG, DenseNet, and Inception. After preprocessing the dataset and deploying these conventional models, the model exhibiting the highest computational efficiency was specified.