What is a reasonably performant algorithm for finding the most "interesting" part of an image?

Question

I have implemented an algorithm described here to highlight the most interesting part of a photo, but it's not very performant, and I'm not sure it does as good a job as some others might. I'd like to try to implement a better one, but I don't actually know enough image processing to get started.

Frankly, I don't even know what questions to ask.

So, what I'm looking for here is a good algorithm that, when provided a jpg image, will find a region of that image that is ... a highlight, if not the highlight, of the image, and ideally did so in a minimum amount of time.

Answer 1

Often this is referred to as salient region detection. There are many ways to do so, as saliency is somewhat objective and mostly depends on the application. By googling you should be able to find many methods. If you give a sample image maybe someone here can provide a more detailed advice. Hope that helps.

Color can often be a good hint for detecting salient parts. The other commonly used approach is to rely on image gradients. These are perhaps the two most generic methods.

Answer 2

You could also have a look at content aware cropping, or seam carving, for those algorithms do the inverse. They try to discard irrelevant or redundant visual content.