Visually impaired people using screen readers will use these textual descriptions to have better understanding of images present in digital contents.
This process of creating textual description for images is a highly time consuming manual process. Moreover, for the same image the textual description may vary from person to person.
Thus, standardizing these Alt-Texts poses a challenge. To address these challenges, we developed Chart-Text, a system that can create textual description given a chart image.
Our proposed Chart-Text system creates complete textual description of five different types of chart images namely: pie charts, horizontal bar charts, vertical bar charts, stacked horizontal bar charts and stacked vertical bar charts. Our system achieves an accuracy of 99.72% in classifying the charts and an accuracy of 78.9% in extracting the data and creating the corresponding Chart-Text.
We built our system by leveraging past research works in areas such as image classification, object detection, localization and optical character recognition. Chart-Text uses both image classification and object detection models, which require a large number of annotated images to train.
To the best of our knowledge, there is no publicly available dataset that satisfies our needs.
We generated images for the different chart types. We randomized the plot aesthetics so that the tool generalizes well for any variances in the data.