Pattern matching methods: Based on Area and Based on features

Question

Good morning, I'm new to image processing and I'm currently working on a project for pattern matching and OCR more specifically.

In the state of art I've found the matching process can be classified into 2: Area based matching and feature based matching. The ABM approaches involve techniques like Cross Correlation which is basically and sliding window, meanwhile the FBM is based on features extraction in order to do the matching of the pattern based on a feature vector...

And here comes my doubt: The extraction of the characteristics can be generally done by algorithms like SIFT and SURF and RANSAC for the selection of the best matching points, but I came across the HOG descriptor which basically would be of good use for pattern matching without having to use any classifier at all, and it would be very simple to implement for the recognition of the characters by just comparing feature vectors.

So my question is, could this HOG descriptor be accurately classified as a method for FBM? OR where in the Computer Vision/Image Processing Area could this be classified? And what would be the best and correct approach to using HOG in the OCR?

Thank you so much in advance!

Answer 1

Yes, if applied properly, HOG is a good feature extractor even for OCR. I could point you towards an OpenCV sample and a MATLAB digit classification tutorial which do exactly that. However, if you like to get state-of-the-art performance, I would suggest deep learning methods, more specifically convolutional neural networks. Many works demonstrated the power of these approaches on MNIST dataset. You might like to check this one or this one for some beginner's tutorials.