Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11915472-B2
Publication Date
2024-02-27
Expiration Date
Abstract
The present disclosure is directed to data classification. An exemplary computer-enabled method for classifying image data comprises: receiving an input image, wherein the input image is of a second data domain; providing the input image to a preprocessing algorithm to obtain a transformed image, wherein the preprocessing algorithm is trained to transform data of the second data domain to data of a first data domain; providing the transformed image to a trained algorithm to analyze the transformed image, wherein the trained algorithm is trained based on training data of the first data domain.
Core Innovation
The invention adapts a pre-trained image classification algorithm that is trained based on training data of a first data domain to analyze images originating from a second data domain. An input image of the second data domain is received and provided to a preprocessing algorithm to obtain a transformed image, and the preprocessing algorithm is trained to transform data of the second data domain to data of the first data domain.
The transformed image is then provided to the trained algorithm to analyze the transformed image. The trained algorithm is trained based on training data of the first data domain, so that the analysis operates on first-domain-like representations produced by the preprocessing algorithm.
In described embodiments, the preprocessing algorithm includes an image style-transfer transformation of second-domain image patches into first-domain-like representations. Training of the preprocessing algorithm uses unpaired two-generator CycleGAN-style translation with adversarial, cycle-consistency, and identity mapping losses, where a first and second image generator and discriminators are used.
Claims Coverage
The independent claims cover four main inventive features: receiving an input image of a second data domain, transforming it via a preprocessing algorithm trained to map the second data domain to a first data domain, analyzing the transformed image using a trained algorithm trained on first data domain training data, and software or storage-medium implementations of the same receive–transform–analyze workflow.
Second-domain input to transformed first-domain-like image
receiving an input image, wherein the input image is of a second data domain; providing the input image to a preprocessing algorithm to obtain a transformed image, wherein the preprocessing algorithm is trained to transform data of the second data domain to data of a first data domain.
First-domain-trained analysis of transformed image
providing the transformed image to a trained algorithm to analyze the transformed image, wherein the trained algorithm is trained based on training data of the first data domain.
Software-implemented receive–transform–analyze electronic device
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for receiving an input image of a second data domain, providing the input image to a preprocessing algorithm to obtain a transformed image trained to transform second data domain data to first data domain data, and providing the transformed image to a trained algorithm to analyze the transformed image trained based on training data of the first data domain.
Non-transitory storage medium with programs for receive–transform–analyze
a non-transitory computer-readable storage medium storing one or more programs comprising instructions which, when executed by one or more processors, cause the electronic device to receive an input image of a second data domain, provide the input image to a preprocessing algorithm trained to transform data of the second data domain to data of the first data domain, and provide the transformed image to a trained algorithm trained based on training data of the first data domain.
The claim coverage is directed to a preprocessing algorithm that maps second data domain images to first data domain representations, followed by analysis using a trained algorithm trained on first-domain training data. The set also includes device and non-transitory medium implementations of the same workflow.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?