Black and white image of document clustering

Question

I have black and white documents (image scanning) and you want to group them according to their layout . To make it more specific, let's say I have the following three images, and the first two are more likely to fall into the same cluster, rather than into the third image, because the first two have a relatively similar layout.

My question is: what would be the best approach to document clustering? Now I have a couple of initial approaches:

• get image hash and compare hash
• using PCA and some clustering methods (K-tool) to compare smaller representations
• extract string using OCR, extract text functions and compare them
• extract a row using OCR and do a keyword search

Will there be other better approaches? Again, only the layout matters.

Answer 1

Do not attempt to copy raw data.

Clustering is uncontrollable; it cannot find out which properties are important and what is not. For the clustering algorithm, everything is important.

Instead, first define the appropriate properties. Such as long edges.