Annotating images is a complex process, and it is easy for us to take for granted that as humans we can ‘see’ and understand what we see. However, when we try to break down this process (or more accurately these processes), we are better able to recognise the assumptions, steps and challenges which may be involved in automating image annotation. On way to illustrate this is by annotating an image ourselves in a structured and systematic way, whereby we need to first identify objects within an image and then apply labels (this is noted previously in the ‘Challenges of annotation’ section). The following article describes the challenges of annotation with specific references to images:

The following article by Adela Barriuso and Antonio Torralba gives an interesting overview of the challenges of image annotation: Notes on image annotation. CoRR abs/1210.3448 (2012)