I'm new to computer vision and I have a common question that I couldn't figure out with Internet or books. As I understood, in general, there are two main approaches in modern computer vision: neural networks based methods and more "classical" methods that use feature extration (edges, keypoints, gradient and etc), like edge detectors, HOG, SIFT and others.

In case of neural networks everything is clear: we have a networks and we use datasets to train it. In other words, we can't use neural networks without previous training it (I call it "pre-training").

And as I understood, we have a bit different situation with features based methods. All examples and techniques that I have already seen consist of two steps. At first, we extract features (maybe with scene pre-processing), and then we use a classifier to recognize the object by those features. If we want classifier to work, we need to train it. So, again, like in case of neural network, we need pre-training.

My question is: are there any computer vision methods that don't need pre-training? Of course, maybe we can use something like a custom, or sometimes "naive" algorithm for certain situation, but are there more common methods? And another related question - are there any methods that can tell us just if the scene contains object, without telling were is it and what kind of that object class is it, again, without pre-training?

Thanks for all answers!


Yup, there are lots of them. For example, image segmentation through graph cuts, camera calibration, image morphing, image stitching, reconstruction of a 3D scene from multiple 2D images, optical flow algorithms, motion tracking, image restoration and inpainting, and probably many more.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.