TARGET DOMAIN CHARACTERIZATION FOR DATA AUGMENTATION

• US 20170300783A1
• Filed: 04/13/2016
• Published: 10/19/2017
• Est. Priority Date: 04/13/2016
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Kumar, Jayant, Xu, Beilei, Paul, Peter
• Granted Patent: US 9,864,931 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/21347 using domain transformations
• CPC Class Codes: G06F 18/2155 characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 10/7753 Incorporation of unlabelled data, e.g. multiple instance learning [MIL]

Abstract

Methods, systems, and processor-readable media for training data augmentation. A source domain and a target domain are provided, and thereafter an operation is performed to augment data in the source domain with transformations utilizing characteristics learned from the target domain. The augmented data is then used to improve image classification accuracy in a new domain.

Claims

1. A method for training data augmentation, said method comprising:

providing a source domain and a target domain;

augmenting data in said source domain with transformations utilizing characteristics learned from said target domain wherein said characteristics include unlabeled images from said target domain; and

employing augmented data from said source domain to said transformations to improve image classification accuracy in said target domain or a new domain.

2. The method of claim 1 wherein said characteristics include image characteristics comprising:

a geometric transformation, a photometric transformation, a resolution conversion, and nuisance transformations.

3. The method of claim 2 further comprising determining at least one representative property of said target domain and said source domain utilizing an average of one property from a set of images and wherein said photometric transformation comprises an intensity clipping.

4. The method of claim 2 further comprising:

deriving source images from said source domain and target images from said target domain;

subjecting said source images and said target images to feature extraction operations; and

comparing results of said feature extraction operation and including said results with said transformations for augmenting said data in said source domain and wherein said resolution conversion comprises a high-resolution to a low-resolution and vice versa.

5. The method of claim 4 wherein said comparing said results further comprises comparing based on at least one of a visual examination and/or an automated process.

6. The method of claim 1 further comprising training a model for said source domain with said augmented data wherein said model comprises a DPM (Deformable Part Model).

7. The method of claim 6 further comprising fine-tuning said model to derive a refined model for said target domain.

8. The method of claim 6 further comprising deriving a new model for said target domain after training said model for said source domain with said augmented data, wherein a SVM (Support Vector Machine) is supplied to said new model for said target domain.

9. The method of claim 4 further comprising:

training a model for said source domain with said augmented data; and

fine-tuning said model to derive a refined model for said target domain.

10. The method of claim 9 further comprising deriving a new model for said target domain after training said model for said source domain with said augmented data.

11. A system for training data augmentation, said system comprising:

at least one processor; and

a non-transitory computer-usable medium embodying computer program code, said computer-usable medium capable of communicating with said at least one processor, said computer program code comprising instructions executable by said at least one processor and configured for;

providing a source domain and a target domain;

augmenting data in said source domain with transformations utilizing characteristics learned from said target domain, wherein said characteristics include unlabeled images from said target domain; and

employing augmented data from said source domain said transformations to improve image classification accuracy in said target domain or a new domain.

12. The system of claim 11 wherein said characteristics include image characteristics comprising:

a geometric transformation, a photometric transformation, a resolution conversion, and nuisance transformations.

13. The system of claim 12 wherein said instructions are further configured for determining at least one representative property of said target domain and said source domain utilizing an average of one property from a set of images and wherein said photometric transformation comprises an intensity chipping.

14. The system of claim 12 wherein said instructions are further configured for:

deriving source images from said source domain and target images from said target domain;

subjecting said source images and said target images to feature extraction operations; and

comparing results of said feature extraction operation and including said results with said transformations for augmenting said data in said source domain and wherein said resolution conversion comprises a high-resolution to a low-resolution and vice versa.

15. The system of claim 14 wherein said comparing said results further comprises comparing based on at least one of a visual examination and/or an automated process.

16. The system of claim 11 wherein said instructions are further configured for training a model for said source domain with said augmented data wherein said model comprises a DPM (Deformable Part Model).

17. The system of claim 16 wherein said instructions are further configured for fine-tuning said model to derive a refined model for said target domain.

18. The system of claim 16 wherein said instructions are further configured for deriving a new model for said target domain after training said model for said source domain with said augmented data wherein a SVM (Support Vector Machine) is supplied to said new model for said target domain.

19. The system of claim 14 wherein said instructions are further configured for:

training a model for said source domain with said augmented data;

fine-tuning said model to derive a refined model for said target domain; and

deriving a new model for said target domain after training said model for said source domain with said augmented data.

20. A system for training data augmentation, said system comprising:

at least one processor; and

a non-transitory computer-usable medium embodying computer program code, said computer-usable medium capable of communicating with said at least one processor, said computer program code comprising instructions executable by said at least one processor and configured for;

providing a source domain and a target domain;

augmenting data in said source domain with transformations utilizing characteristics learned from said target domain, wherein said characteristics include unlabeled images from said target domain;

employing augmented data from said source domain said transformations to improve image classification accuracy in a new domain;

training a model for said source domain with said augmented data;

fine-tuning said model to derive a refined model for said target domain; and

deriving a new model for said target domain after training said model for said source domain with said augmented data wherein a SVM (Support Vector Machine) is supplied to said new model for said target domain.