公开(公告)号：US11341605B1

公开(公告)日：2022-05-24

申请号：US16588503

申请日：2019-09-30

Applicant: Amazon Technologies, Inc.

Inventor： Kunwar Yashraj Singh ,  Amit Adam ,  Shahar Tsiper ,  Gal Sabina Star ,  Roee Litman ,  Hadar Averbuch Elor ,  Vijay Mahadevan ,  Rahul Bhotika ,  Shai Mazor ,  Mohammed El Hamalawi

IPC: G06T3/00 ,  G06T1/20 ,  G06T3/60 ,  H04L67/02

Abstract: Techniques for document rectification via homography recovery using machine learning are described. An image rectification system can intelligently make use of multiple pipelines for rectifying document images based on the detected type of device that generated the images. The image rectification system can provide high-quality rectifications without requiring human cooperation, multiple views of the document in multiple images, and/or without being constrained to only be able to process images from one source context.

公开(公告)号：US11087081B1

公开(公告)日：2021-08-10

申请号：US16359930

申请日：2019-03-20

Applicant: Amazon Technologies, Inc.

Inventor： Amulya Srivastava ,  Vivek Bhadauria ,  Gowtham Jeyabalan ,  Paul H. Kang ,  Mohammed El Hamalawi

IPC: G06F40/00 ,  G06F40/186 ,  G06N3/04 ,  G06N20/00 ,  G06F40/117 ,  G06F40/169

Abstract: A synthetic document generator that obtains a configuration for a synthetic document derived from real-world documents. The configuration specifies element templates to be included in the synthetic document and weights for the specified element templates. The system generates synthetic documents based on the configuration; the synthetic documents include diversified versions of the element templates specified in the configuration. Annotation documents are generated for the synthetic documents that include information describing the respective synthetic documents. A machine learning model for analyzing real-world documents can then be trained using the synthetic and annotation documents. Feedback from the analysis of real-world documents by the machine learning model can be used to generate a new configuration for generating additional synthetic and annotation documents which are used to further train the model.