ML on Images: Best Practices?

Is there a "how to do Machine Learning on image data: for dummies"-style guide out there? Or a "do this / never do this" best practices kind of guide?

To put in context: I have experience with ML and data, just never ML on images. Besides "use the histogram" and bag-of-words, never gotten any other advice.

Thanks for any tips!

EDIT:

I'm interested in image classification/categorization and object recognition.