Microsoft estimates that more than half of the images circulating on the Internet have no captions, which is irrelevant information for those who can see the content, but critical for giving oral context to those who cannot see or who have a reduced level of vision.
To minimize the impact of this lack of information, the company has updated Edge with a new feature, which will automatically create a description for images that don’t have it. This textual description will now be available to be read by screen readers, who until now the only information they could return, when a description was not detailed by the person who created the image, was a message that the chart has no legend.
Now, “when a screen reader encounters an image without a caption, that image can be automatically processed by machine learning algorithms to describe the image in words and capture any text it may contain,” Microsoft says. .
The company admits that “the algorithms aren’t perfect and the quality of the description will vary, but for those using screen readers, having some description of an image will be much better than having no context at all.” at all”, defends Travis Leithead, product manager of the Microsoft Edge browser, quoted in a note on the site. On the other hand, and since it uses artificial intelligence, the functionality will improve over time, as the task is repeated and the diversity of the images analyzed increases.
The captioning work will be done by a computer vision API on the Azure platform, which will process the information from each image to create alternative text, which will be associated with the features of the image. This written account of visual content will be available in Portuguese, English, Spanish, Japanese and Simplified Chinese.
The update takes effect immediately for all desktop versions of Edge, i.e. Windows, Mac, and Linux. It can be enabled in the accessibility options of browser settings and used with any screen reader. Anyone using the browser on Android or iOS will not be able, at least for now, to take advantage of the novelty. Also excluded from this effort are images that the API identifies as sexual, obscene, or decorative, as well as very large or very small images (less than 50×50 pixels).