Assuming imagination can be roughly defined as "taking things from reality/experience and mixing them up somehow to create something dissimilar from the inputs", then any cognitive framework:
- with a probabilistic interpretation
- who is capable of analogy
roughly models how imagination may manifest itself in the human mind.
Imagination as inference
Imagine you've ridden two bicycles: a fast bicycle and a slow bicycle. The bicycle salesman now shows you a third bicycle with a speed in between the two you've previously ridden. Although you do not have any direct experience with the bicycle, you can infer from your two previous experiences that the bicycle will give you an experience somewhere in between the two. This is what I mean by "probabilistic interpretation". If your cognitive model has a concept of a distribution of possible bicycle speeds, it can infer points along that distribution. This approach is used in cognitive models like neural networks.
Imagination as analogy
Consider the analogy:
Structure of an atom is like a solar system. Nucleus is the sun and
electrons are the planets revolving around their sun.
Given basic knowledge of the solar system, despite never seeing atoms, you can imagine a few things about atom structure. You can imagine electrons float freely around the center of the nucleus without slowing down. You were able to combine a symbol you had for the solar system with the unknown symbols of electrons and nuclei. This type of imagination is found in cognitive models like ACT-R.
Imagination as something more complicated
The two previous forms of imagination were pretty simple in terms of interactions. Two inputs formed a novel output, but scientists have all sort of theories about how we build models in our heads that involve scaling up to more than one input, drawing experiences from memory and interacting with the environment. There are so many that you have to start developing criteria for what the best model of imagination is. Obviously, I can't cover all the models, but hopefully this gives you a good starting point.