“ZoIA”: uma ferramenta para descrição automática de imagens no navegador.

Abstract

The proper functioning of screen readers, commonly used by visually impaired users, is linked to the responsibility of web developers to use tags correctly and provide sufficient information, such as image descriptions, to ensure accessibility and inclusion for these users. However, the provision of alternative texts is not always guaranteed due to several factors, such as the absence of descripti-ons associated with images, as well as requirements and functionalities that seek to guarantee the existence of this data. In view of this problem, this work proposes “ZoIA”, a browser plugin, based on Artificial Intelligence, that describes and inserts alternative texts for images displayed on web pages. A comparative analysis between the direct generation of descriptions in Portuguese and the sequence based approach suggested that adjusting the model to the language of interest can result in qualitative improvements in the context, semantics and grammatical cohesion of automatic descriptions. This work sought to contribute to improving the interaction and experience of visually impaired people when using web applications using state-of-the-art Artificial Intelligence strategies.