This contain the sketch of a receipt_recognition_module, a module used to extract data (only total in euro for now) from receipts, and some test on this module. It take a image from the image module, transform this image into a matrix and effectuate transformation (cropping, binarization, line segmentation) then use a LSTM (Long Short-Term Memory) neural network written in python (from ocropy library) to recognize the characters.
It is divided in 2 commit: one containing the code that execute the recognition, the second containing the tests and images that are tested. This merge request is dependant on this one: https://lab.nexedi.com/nexedi/slapos/merge_requests/153