Detecting 'anomalies' in structured / textured image

asked 2016-08-08 09:59:26 -0500

Nbb gravatar image

updated 2016-08-08 10:25:52 -0500

Given a textured or a structured image, how can I localize anomalous segments ? For example a large crack or graffiti or just things that do not match well to the pattern (even a window like in the 3rd picture).

I thought of using SLIC superpixels as the first stage but I was wondering if someone can post a link to a paper that discusses a similar problem. I am planning to make use of training data.

image description

image description

image description

edit retag flag offensive close merge delete


This is a very interesting topic. I'm working on a project and want to develop ML algorithm to find anomalies on the same ''pattern'' sufraces as You asked. Our idea (not working yet) is to separate pattern image on the very small portions that would be used as training data for neural nets. There are many problems that we have including not having enought of training data...

jok23 gravatar imagejok23 ( 2016-11-16 12:27:19 -0500 )edit