Evaluating the Effectiveness of STEM Images Captioning
Maurizio Leotta, Marina Ribaudo · 2024
Despite the advent of the World Wide Web over three decades ago and the prevalence of online activities in contemporary society, barriers for users with disabilities to access online content still exist.With the proliferation of smartphones and easy access to cameras, the number of images shared daily on social media platforms is substantial and continuously growing.For visually impaired users, this abundance of images can be perceived as a significant obstacle to the understanding of online content.This study presents the initial findings of a long-term experiment focused on exploring the challenges in creating alternative texts for complex images, particularly those pertinent to disciplines encompassing mathematics, physics, technology, and the natural sciences.Our investigation is focused on two main aspects.Firstly, we explore the feasibility of using AI engines to generate alternative descriptions for this type of images.Secondly, we assess how basic training on STEM accessibility impacts undergraduate students' awareness.The results indicate that conveying the meaning of complex images to visually impaired users through descriptions remains challenging, both for students and AI engines.However, it is encouraging to note that even basic training has empowered students to write improved descriptions and to critically assess those composed by their peers.This underscores the significance of education in promoting a foundational awareness among future software developers concerning the crucial yet frequently overlooked aspect of web accessibility.