Image rotation
AssigneeCanva Pty Ltd
InventorsSanchit Sanchit, Stefan Sietzen
Life of this patent
Estimating the term…
Abstract
Methods of training a machine learning model for image processing are described, as well as methods for image processing to determine a predicted rotation value for an image based on the trained machine learning model. A method of training includes utilising as a learning objective a reduction or minimisation of a classification loss. Training image pairs may be associated with classifications corresponding to rotation values for the purpose of determining the classification loss. An image may be divided into a plurality of images for determining an input to the machine learning model, for training or for image processing.
Claims20
A computer-implemented method including:
- determining, via one or more computer processors, that an image requires rotation correction by applying a trained classifier model to the image, wherein the trained classifier model is configured to determine whether the image contains rotation information;
- dividing, via one or more computer processors, the image into a plurality of images, including at least a first image corresponding to a first part of the image and a second image corresponding to a second part of the image, different to the first part of the image;
- forming, via the one or more computer processors, a first feature vector for the first image and a second feature vector for the second image;
- concatenating, via the one or more computer processors, into a single vector the first feature vector and the second feature vector and providing, via the one or more computer processors, the single vector as an input to the machine learning model, wherein the machine learning model is configured to predict a rotation of the image based on the feature vector;
- determining, via the one or more computer processors, that a predicted rotation class output by the machine learning model is within a predetermined range of rotation values;
- and
- applying, via the one or more computer processors, the predicted rotation to rotate the image based on the determination that the predicted rotation class is within the predetermined range.
A computer processing system including one or more computer processors and computer-readable storage, the computer processing system configured to perform a method including:
- determining, via one or more computer processors, that an image requires rotation correction by applying a trained classifier model to the image, wherein the trained classifier model is configured to determine whether the image contains rotation information;
- dividing, via the one or more computer processors, the image into a plurality of images, including at least a first image corresponding to a first part of the image and a second image corresponding to a second part of the image, different to the first part of the image;
- forming, via the one or more computer processors, a first feature vector for the first image and a second feature vector for the second image;
- concatenating, via the one or more computer processors, into a single vector the first feature vector and the second feature vector and providing, via the one or more computer processors, the single vector as an input to the machine learning model, wherein the machine learning model is configured to predict a rotation of the image based on the feature vector
- determining, via the one or more computer processors, that a predicted rotation class output by the machine learning model is within a predetermined range of rotation values;
- and
- applying, via the one or more computer processors, the predicted rotation to rotate the image based on the determination that the predicted rotation class is within the predetermined range.
Non-transitory computer-readable storage storing instructions for a computer processing system, wherein the instructions, when executed by the computer processing system cause the computer processing system to perform a method including:
- determining, via one or more computer processors, that an image requires rotation correction by applying a trained classifier model to the image, wherein the trained classifier model is configured to determine whether the image contains rotation information;
- dividing, via one or more computer processors, the image into a plurality of images, including at least a first image corresponding to a first part of the image and a second image corresponding to a second part of the image, different to the first part of the image;
- forming, via the one or more computer processors, a first feature vector for the first image and a second feature vector for the second image;
- concatenating, via the one or more computer processors, into a single vector the first feature vector and the second feature vector and providing, via the one or more computer processors, the single vector as an input to the machine learning model, wherein the machine learning model is configured to predict a rotation of the image based on the feature vector;
- determining, via the one or more computer processors, that a predicted rotation class output by the machine learning model is within a predetermined range of rotation values;
- and
- applying, via the one or more computer processors, the predicted rotation to rotate the image based on the determination that the predicted rotation class is within the predetermined range.
Dependencies are inferred from the claim text.