Generating images using image assets extracted from other images
AssigneeAmazon Technologies, Inc.
InventorsSteve Huynh, Erik Martin Zigman, Ronald Diaz
Life of this patent
Estimating the term…
Abstract
Systems, devices, and methods are provided for processing images using machine learning. Features may be obtained from an image using a residual network, such as ResNet-101. Features may be analyzed using a classification model such as K-nearest neighbors (K-NN). Features and metadata extracted from images may be used to generate other images. Templates may be used to generate various types of images. For example, assets from two images may be combined to create a third image.
Claims20
A computer-implemented method, comprising:
- obtaining a first image;
- extracting, using a residual network, a first plurality of features from the first image;
- determining, using a K-nearest neighbors (K-NN) classifier, first metadata associated with a first feature of the first plurality of features;
- obtaining a second image;
- extracting, using the residual network, a second plurality of features from the second image;
- determining, using the K-NN classifier, second metadata associated with a second feature of the second plurality of features;
- obtaining a template for a third image, wherein the templates comprises:
- a first element that encodes third metadata for first content to be generated at a first portion of the third image;
- and
- a second element that encodes fourth metadata for second content to be generated at a second portion of the third image;
- determining the first metadata matches the third metadata;
- determining the second metadata matches the fourth metadata;
- and
- generating the third image based on the template, wherein the first portion of the third image is generated based on the first feature and the second portion of the third image is generated based on the second feature.
A system, comprising:
- one or more processors;
- and
- memory storing executable instructions that, as a result of execution by the one or more processors, cause the system to:
- obtain a first image;
- determine, using one or more convolutional neural networks, a set of features in the first image;
- determine, using one or more classification models, a first set of properties for a first feature of the set of features;
- obtain a template for a second image, wherein the template comprises a first element and a second set of properties;
- determine, based on the first set of properties and the second set of properties, that the first feature is suitable to be used as the first element;
- and
- generate, based on the first feature, a portion of the second image.
A non-transitory computer-readable storage medium storing executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to at least:
- obtain a set of images;
- determine, using one or more convolutional neural networks, a first feature for a first image of the set of images;
- determine, using one or more classification models, first metadata of the first feature;
- determine second metadata for a first portion of a second image;
- determine, based on the first metadata and the second metadata, that the first feature is suitable for use in the first portion of the second image;
- and
- generate, using the first feature, the first portion of the second image.
Dependencies are inferred from the claim text.