Sample Python TensorFlow for Classifying Document Images

I'd like a basic example of a method for classifying and grouping document images into known categories. We would start with a known genre or layout-type of documents based on images of known documents. Then compare each new document with those layouts. If they are close in appearance to one or more layouts, the new documents are classified accordingly. For example, if we have a number of known document "shapes", which are described by the visual arrangement of items on the page, we should be able to load or train the system to understand those shape->category relationships. Then the system could classify the images into categories for downstream processing. Useful images for the simple demonstration would be a Driver's License, a bank check, a job application. On another subsequent project we would consider using textual features, but this project is just based on images/contours. Some reference: general: [login to view URL] edge detection: [login to view URL] [login to view URL] [login to view URL]