Ameesh, I think it's a little more complicated than it appears from the abstract. You have to dig down into the claims a bit.
Here's how it might work.
The user responds to a query which includes questions like name, sex, age, hobbies, etc.
Based on the user input, an image grouping and annotaion grouping is slected. Say for example, the user selects sport under hobbies and then further selects basketball. A futher slection could be favorite team.
The image grouping would then be the Knicks, for example, and the annotation grouping would relate to basketball.
The user than inputs his image. The image may be imported via a variety of means.
The software then creates a unique personalized image which is a combination of an annotation, a predefined image, and the user inputed image. The composite image is random within the groupings determined by the query.
For example the composite image could contain Alan Huston's body with the users head superimposed and a caption such as "great shot, user's name, nothing but net" or something like that.
This is a corny example, but hopefully you get the idea.
If the user doesn't like the composite image, he can run the random generator again and again until he get's an image he likes.
The advantage is that outcome has an element of surprise with minimal user intervention.