Please use this identifier to cite or link to this item:

http://hdl.handle.net/10609/73186
Title: Classification of identity documents using a deep convolutional neural network
Author: Vilàs Mari, Pere
Director: Meler Corretjé, Lourdes
Tutor: García Solórzano, David  
Morán Moreno, José Antonio
Others: Universitat Oberta de Catalunya
Keywords: machine learning
computer vision
identity document
Issue Date: 20-Jan-2018
Publisher: Universitat Oberta de Catalunya
Abstract: In this work, we review the main subjects that we have to serve to build an image classifier, then we design and implement a system to classify identity documents. Especially relevant is the description of the workflow that we have come up with after several ways of approaching the problem and which we hope can serve other machine learning practitioners. We evaluate some fea- ture extractor algorithms to find the most suitable for identity documents classifi- cation purposes. Using virtual machines on the cloud, we run feature extractors in parallel to label at a speed of 16 images/s. Then, we select a neural network architecture and hyperparameters to train a convolutional neural network. Our results give an accuracy of 98%. We detail the responses of the convolutional filters over the images. Results and source code in the appendix.
Language: English
URI: http://hdl.handle.net/10609/73186
Appears in Collections:Bachelor thesis, research projects, etc.

Share:
Export:
Files in This Item:
File Description SizeFormat 
Presentacio-identity-doc-classificator.mp4Video de presentació del treball154.24 MBMP4View/Open
pvilasTFM0118memoria.pdfMemoria del TFM7.89 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons