Images annotated according to their content: a study on the description of data in image format in multiple domains

When it comes to images, the challenge is increased, as they have characteristics that differentiate them from other typologies. We conducted a study in which we obtained a set of 27 images described according to their content, by researchers of the projects where they are inserted.

After obtaining the ground-truth that would support the analysis, we proceeded to two more stages of description, one through an automatic processing tool (Vision AI) and the other through researchers with no knowledge of the images. We concluded that the human description is more elucidative of the images' content, namely at a semantic level. In turn, the automatic tools enhance a more literal description.