AIDE: automatic and accessible image descriptions for review imagery in online retail

Abstract

Review imagery is a critical way to share products in authentic contexts. This visual method of showcasing a product can help sighted people govern their purchasing decisions, as it can include details not apparent in the product page and image. However, visually impaired people are not afforded the same opportunity due to lack of image descriptions. Previous research has found that context impacts the desired description; our studies have identified key features to extract from review imagery in the online retail context. In this paper, we present AIDE, an Automatic Image Description Engine that leverages computer vision and natural language processing to automatically generate image descriptions for user-generated imagery. Results from our evaluative studies show that AIDE encourages online retail engagement, fosters an inclusive shopping environment, and increases the feeling of independence.