Ethiopian doctoral student creates first public dataset for African literacy

Wondimu Gebre Dikubab, an Ethiopian doctoral student studying in China at Huazhong University of Science and Technology, released the first comprehensive public dataset and preliminary algorithm solution for Amharic scene text detection and recognition.

The first comprehensive and public Amharic text recognition dataset, released with the help of Professor Bai Xiang, is composed of four parts: HUST-ART, HUST-AST, A.B.E., and Tana, with a total of 15,039 real scene text images and 2,927,682 composite text images.

Dikubab has designed several experiments to evaluate the performance of current state-of-the-art text recognition methods in recognizing Amharic text on this dataset.

Entitled Comprehensive Benchmark Datasets for Amharic Scene Text Detection and Recognition, the research results were published in SCIENCE CHINA Information Sciences.

The work is a milestone for the technological development of Amharic and will directly contribute to the application of text recognition in the official Ethiopian language and other African languages.

Amharic is the official language of Ethiopia, which is an active participant in the Belt and Road Initiative, and the second-largest Semitic language family after Arabic, with a wide range of applications worldwide.

Detection and recognition based on Amharic text and pictures will contribute to the digital and intelligent transformation of east African countries in administration, transportation, tourism, and other related fields.

Huazhong University of Science and Technology has trained more than 10,000 international students from more than 150 countries since 1962, making positive contributions to promoting cultural exchanges, enhancing friendship among different people, and facilitating global cooperation.