Method for embedding information in a decorative label

Patent metadata

Field	Value
Publication number	US-12555175-B2
Application number	US-202117922714-A
Country	US
Kind code	B2
Filing date	May 5, 2021
Priority date	May 6, 2020
Publication date	Feb 17, 2026
Grant date	Feb 17, 2026

Abstract

A computer-implemented method for creating an encoder-decoder system for embedding information in a decorative label. The method includes defining a family of encoder functions and a family of decoder functions. Each encoder function of the family of encoder functions configured to encode information as a respective modification of a decorative label. Each decoder function of the family of decoder functions configured to decode an image of a modified decorative label into respective decoded information. The method includes applying an iterative optimization process to determine an optimized encoder-decoder pair. The optimized encoder-decoder air includes an optimized encoder function and an optimized decoder function. The optimized encoder function is selected by the iterative optimization process from the family of encoder functions and the optimized decoder function is selected by the iterative optimization process from the family of decoder functions.

First claim

What is claimed is:

1 . A computer-implemented method for creating an encoder-decoder system for embedding information in a label, the method comprising:
receiving an encoder specification indicative of one or more modification rules, each modification rule defining a type and/or magnitude of a modification to the label; defining a family of encoder functions and a family of decoder functions, each encoder function of the family of encoder functions configured to encode information as a respective modification of a label, each decoder function of the family of decoder functions configured to decode an image of a modified label into respective decoded information; and applying an iterative optimization process to determine an optimized encoder-decoder pair, the optimized encoder-decoder pair comprising an optimized encoder function and an optimized decoder function, the optimized encoder function selected by the iterative optimization process from the family of encoder functions and the optimized decoder function selected by the iterative optimization process from the family of decoder functions, the iterative optimization process being configured to determine the optimized encoder-decoder pair so as to at least approximately optimize a detection reliability of the optimized decoder function when decoding images of labels that have been modified using the optimized encoder function, wherein: each encoder function of the family of encoder functions is configured to receive a vector graphics representation of the label and to output a modification of the vector graphics representation; wherein the family of encoder functions is a parametric family of encoder functions, parametrized by one or more adaptable encoder parameters; wherein the family of decoder functions is a parametric family of decoder functions, parametrized by one or more adaptable decoder parameters, and wherein the iterative optimization process comprises iteratively adapting the one or more encoder parameters and the one or more decoder parameters; wherein the vector graphics representation includes a plurality of vector graphic primitives, each primitive having one or more properties; and wherein the modification of the vector graphics representation includes a modification of one or more of the primitives; and wherein the method comprises receiving an encoder specification indicative of one or more modification rules, each modification rule being specific to a particular label and defining a type and/or magnitude of a modification to specific primitives of the label.

2 . A method according to claim 1 , wherein the optimized encoder-decoder pair is determined using a machine-learning process based on a set of training examples, each training example including a training label and training information to be embedded in the training label.

3 . A method according to claim 2 , wherein the family of decoder functions is defined as a neural network, in particular as a convolutional neural network, parametrized by a plurality of adaptable network weights.

4 . A method according to claim 2 , wherein applying the modification to a label during use of the optimized encoder function comprises a rasterization of the modified vector graphics representation resulting in rasterized version of the modified vector graphics representation, and wherein the optimization process is configured to apply a differentiable approximation of said rasterization process resulting in a smoothed rasterized version of the modified vector graphics representation.

5 . A method according to claim 1 , wherein the iterative optimization process comprises a first optimization stage and a second optimization stage, wherein, during the first optimization stage, the encoder parameters are kept constant, and only the decoder parameters are adapted according to an optimization update rule responsive to a suitable error measure, and wherein, during the second optimization stage, the encoder parameters and the: decoder parameters are adapted according to the same or a different optimization update rule.

6 . A method according to claim 1 , wherein:
the encoder-decoder pair is optimized for use with one or more particular user-defined labels, and the set of training examples comprises the one or more user-defined labels, one or more corresponding encoder specifications, each indicative of one or more modification rules specific to a respective one of the one or more user-defined labels, and a plurality of items of training information to be embedded in the one or more user-defined labels.

7 . A method according to claim 6 , wherein the set of training examples comprises only a single user-defined label.

8 . A method according to claim 1 , wherein:
the vector graphics representation of the label comprises one or more geometric primitives, each geometric primitive having one or more associated properties, and each encoder function of the family of encoder functions is configured to compute the modification of the vector graphics representation as a set of modifications of respective values of one or more of said properties, the set of modifications encoding said information.

9 . A method according to claim 8 , wherein:
said properties are represented by respective continuous variables, and each encoder function of the family of encoder functions is configured to compute the modification of the vector graphics representation as a set of modifications of respective values of one or more of said continuous variables, the set of modifications encoding said information.

10 . A method according to claim 1 , further comprising:
a) receiving a label and at least one item of information to be embedded into said label, b) using the optimized encoder function to encode the received item of information as a modification of the received label, and c) applying the modification to the received label to obtain a modified label.

11 . A method according to claim 10 , wherein the label is a user-defined label and the encoder-decoder system is optimized for the particular user-defined label.

12 . A method according to claim 1 , further comprising:
a) obtaining one or more images of respective modified labels, and b) applying the optimized decoder function to decode each of the obtained images of the respective modified labels into respective decoded information.

13 . A method for embedding information in a label, the method comprising the steps of: obtaining, by a data processing system executed on a computer processor, the label in a vector graphics format having geometric primitives, receiving, by the data processing system, modification rules defining metes and bounds of acceptable alterations to the label, and executing, by the data processing system, a training process configured to train an encoder to embed information into the label by altering the geometric primitives of the label while observing the modification rules to form a modified label, the training process also configured to train a decoder to detect the information embedded in the modified label, the training process configured to iteratively adapt both the encoder and decoder for optimization; wherein the vector graphics representation includes a plurality of vector graphic primitives, each primitive having one or more properties; and wherein the modification of the vector graphics representation includes a modification of one or more of the primitives, wherein the method comprises receiving an encoder specification indicative of one or more modification rules, each modification rule being specific to a particular label and defining a type and/or magnitu

Assignees

• Lego As

Inventors

• Pedersen Henrik
• Noe Karsten Østergaard
• Hjermitslev Oliver Gyldenberg
• Rasmussen Søren
• Døssing Philip Kongsgaard
• Oliver Simon
• Harr Rasmus
• Prahm Mads
• Ignatov Vladimir
• Søderberg Jesper

Classifications

• G06T2201/0053

  Embedding of the watermark in the coding stream, possibly without decoding; Embedding of the watermark in the compressed domain · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/045

  Combinations of networks · CPC title