Imagery contents descriptions for People with visual impairments

Abstract

Image descriptions are crucial in assisting individuals without eyesight by providing verbal representations of visual content. While manual and Artificial Intelligence (AI)-generated descriptions exist, automatic description generators have not fully met the needs of visually impaired People. In this study, we have examined the problems related to image descriptions reported in existing literature using the Snowballing technique. Through this method, we have identified thirteen issues, including ethical concerns surrounding physical appearance, gender and identity, race, and disability. Furthermore, we have identified five reasons why sighted Individuals often fail to provide descriptions for visual content, highlighting the necessity for accessibility campaigns that raise awareness about the social significance of descriptive sentences. We conducted interviews with eight low-vision volunteers, in which we analyzed the characteristics of descriptive sentences for 25 indoor images and gathered participants’ expectations regarding image descriptions. As a result, we propose a set of Good Practices for writing descriptive sentences aimed to assist automatic tools and sighted Individuals in generating more satisfactory and high-quality image descriptions. We hope our results will emphasize the societal importance of imagery descriptions and inspire the community to pursue further interdisciplinary research to address the issues identified in our study.

Cite this work

Stored copy: cc-by-nc-sa · source https://hdl.handle.net/10923/19442. Open access only — a takedown request can remove the copy without touching this record.

Stored only when both papers are in the corpus — 4 citations exist upstream.