Deep learning / computer vision technique: aggregating many input images to a single representation of the features within

I have a few thousand grayscale images, and I would like to generate a universal representation of the patterns within - a semantic/ordered composition of all features, so to speak. For instance, take 10000 images of a dog and draw the archetypical dog.

Does this task have a technical name, and is there a method out there specifically for such purposes?

I guess this similar to what happens during the training of a neural network. I just don't necessarily need a model for prediction afterwards (don't mind if I do though), just the aggregate representation will do.

The images have different sizes, and the patterns in there are not scale and rotation invariant, so simple averaging algorithms wont work.