Image Descriptions' Limitations for People with Visual Impairments: Where Are We and Where Are We Going?

Image descriptions aim to transcribe the visual content and are essential for people who do not have eyesight. Such image descriptions are generated manually or by Artificial Intelligence (AI) models. Despite its relevance, the emergence of automatic image descriptions was not motivated by people with visual impairments; thus, they still cause dissatisfaction in this audience. This paper provides a snowballing review of the limitations of image descriptions for people with visual impairments. We encountered thirteen image description issues, including those analogous to the ethics, such as people's appearances, gender, race, and disability. We identified five reasons why sighted people do not write image descriptions for the content they share, exposing the necessity of accessibility campaigns to awareness people of the social importance of image descriptions. Moreover, we discussed recommendations found in the literature that may support automated tools and sighted people for writing high-quality image descriptions. We hope our results will highlight the social significance of image descriptions and encourage the community to pursue further interdisciplinary researches that could potentially minimize the issues pointed out in our study.