Methods and systems for encoding images

Patent metadata

Field	Value
Publication number	US-12175740-B2
Application number	US-201917614929-A
Country	US
Kind code	B2
Filing date	May 28, 2019
Priority date	May 28, 2019
Publication date	Dec 24, 2024
Grant date	Dec 24, 2024

Abstract

The present disclosure is directed to encoding images. In particular, one or more computing devices can receive data representing one or more machine learning (ML) models configured, at least in part, to encode images comprising objects of a particular type. The computing device(s) can receive data representing an image comprising one or more objects of the particular type. The computing device(s) can generate, based at least in part on the data representing the image and the data representing the ML model(s), data representing an encoded version of the image that alters at least a portion of the image comprising the object(s) such that when the encoded version of the image is decoded, the object(s) are unrecognizable as being of the particular type by one or more object-recognition ML models based at least in part upon which the ML model(s) configured to encode the images were trained.

First claim

What is claimed is:

1. A computer-implemented method comprising:
receiving, by one or more computing devices, data representing one or more machine learning (ML) models configured, at least in part, to encode images comprising objects of a particular type;
receiving, by the one or more computing devices, data representing an image comprising one or more objects of the particular type;
generating, by the one or more computing devices and based at least in part on the data representing the image and the data representing the one or more ML models, data representing an encoded version of the image that alters at least a portion of the image comprising the one or more objects such that when the encoded version of the image is decoded, the one or more objects are unrecognizable as being of the particular type by one or more object-recognition ML models based at least in part upon which the one or more ML models configured to encode the images were trained; and
communicating, by the one or more computing devices and to a remotely located computing system, the data representing the encoded version of the image, wherein the remotely located computing system is configured to:
receive the data representing the encoded version of the image; and
generate, based at least in part on the data representing the encoded version of the image and one or more ML models trained in conjunction with the one or more ML models configured to encode the images, data representing a decoded version of the image in which the one or more objects are unrecognizable as being of the particular type by the one or more object-recognition ML models.

2. The computer-implemented method of claim 1 , wherein:
generating the data representing the encoded version of the image comprises generating data representing the encoded version of the image such that one or more objects of a different type from the particular type are recognizable in the decoded version of the image as being of the different type by at least one of the one or more object-recognition ML models, and
the remotely located computing system is configured to utilize the at least one of the one or more object-recognition ML models to identify the one or more objects of the different type in the decoded version of the image as being of the different type.

3. A system comprising:
one or more processors; and
a memory storing instructions that when executed by the one or more processors cause the system to perform operations comprising:
receiving, from a remotely located computing device, data representing an encoded version of an image comprising one or more objects of a particular type, the encoded version of the image altering at least a portion of the image comprising the one or more objects such that when the encoded version of the image is decoded, the one or more objects are unrecognizable as being of the particular type by one or more object-recognition machine learning (ML) models;
identifying data representing one or more ML models generated based at least in part on the one or more object-recognition ML models;
generating, based at least in part on the data representing the encoded version of the image and the data representing the one or more ML models generated based at least in part on the one or more object-recognition ML models, data representing a decoded version of the image in which the one or more objects are unrecognizable as being of the particular type by the one or more object-recognition ML models, wherein generating the data representing the decoded version of the image comprises generating data representing the decoded version of the image such that one or more objects of a different type from the particular type are recognizable in the decoded version of the image as being of the different type by at least one of the one or more object-recognition ML models; and
utilizing the at least one of the one or more object-recognition ML models to identify the one or more objects of the different type in the decoded version of the image as being of the different type.

4. The system of claim 3 , wherein:
the image comprises one or more objects of a second different type from the particular type;
the data representing the encoded version of the image comprises data that alters at least a portion of the image comprising the one or more objects of the second different type such that when the encoded version of the image is decoded, the one or more objects of the second different type are unrecognizable as being of the second different type by the one or more object-recognition ML models; and
the one or more objects of the second different type are unrecognizable, in the decoded version of the image, as being of the second different type by the one or more object-recognition ML models.

5. The system of claim 4 , wherein the data representing the encoded version of the image comprises data generated based at least in part on a common ML model configured to encode the image such that the encoded version of the image alters:
the at least a portion of the image comprising the one or more objects of the particular type such that when the encoded version of the image is decoded, the one or more objects of the particular type are unrecognizable as being of the particular type by the one or more object-recognition ML models; and
the at least a portion of the image comprising the one or more objects of the second different type such that when the encoded version of the image is decoded, the one or more objects of the second different type are unrecognizable as being of the second different type by the one or more object-recognition ML models.

6. The system of claim 4 , wherein the data representing the encoded version of the image comprises data generated based at least in part on:
a first ML model configured to encode the image to generate data representing a modified version of the image such that the modified version of the image alters the at least a portion of the image comprising the one or more objects of the particular type such that when the modified version of the image is decoded, the one or more objects of the particular type are unrecognizable as being of the particular type by the one or more object-recognition ML models; and
a second ML model configured to encode a decoded version of the modified version of the image to generate data representing the encoded version of the image such that the encoded version of the image alters the at least a portion of the image comprising the one or more objects of the second different type such that when the encoded version of the image is decoded, the one or more objects of the second different type are unrecognizable as being of the second different type by the one or more object-recognition ML models, the second ML model being different and distinct from the first ML model.

7. The system of claim 3 , wherein the operations comprise utilizing one or more ML models to reconstruct, in the decoded version of the image, the at least a portion of the image comprising the one or more objects such that a human viewing a rendering of the decoded version of the image can identify, in the rendering, the one or more objects as being of the particular type.

8. One or more non-transitory computer-readable media comprising instructions that when executed by one or more computing devices cause the one or more computing devices to perform operations comprising:
receiving data representing a plurality of different and distinct images that includes at least one image comprising an object of a particular type; and
generating, based at least in part on the data representing the plurality of different and distinct images and one or more object-recognition machine learning (ML) models,

Assignees

• Google Llc

Inventors

• Baluja Shumeet
• Sukthankar Rahul

Classifications

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/094

  Adversarial learning · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0455

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