Method and computer program product for producing a pattern recognition training set
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06F-017/50
G06F-015/18
G06K-009/46
출원번호
US-0105714
(2002-03-25)
발명자
/ 주소
Ii,David L.
Reitz, II,Elliott D.
Tillotson,Dennis A.
출원인 / 주소
Lockheed Martin Corporation
대리인 / 주소
Tarolli, Sundheim, Covell&
인용정보
피인용 횟수 :
5인용 특허 :
24
초록▼
The present invention recites a method and computer program product for generating a set of training samples from a single ideal pattern for each output class of a pattern recognition classifier. A system equivalent pattern is generated for each of a plurality of classes from a corresponding ideal p
The present invention recites a method and computer program product for generating a set of training samples from a single ideal pattern for each output class of a pattern recognition classifier. A system equivalent pattern is generated for each of a plurality of classes from a corresponding ideal pattern. A noise model, simulating at least one type of noise expected in a real-world classifier input pattern, is then applied to each system equivalent pattern a set number times to produce, for each output class, a number of training samples. Each training sample simulates defects expected in real-world classifier input patterns.
대표청구항▼
Having described the invention, we claim: 1. A method for generating a set of X training samples from a single ideal pattern for each output class of a pattern recognition classifier, comprising: generating a system equivalent pattern for each of a plurality of classes from a corresponding ideal pa
Having described the invention, we claim: 1. A method for generating a set of X training samples from a single ideal pattern for each output class of a pattern recognition classifier, comprising: generating a system equivalent pattern for each of a plurality of classes from a corresponding ideal pattern; and applying a noise model, simulating a plurality of noise types expected in a real-world classifier input pattern, where each of the plurality of noise types are defined by at least one of a plurality of parameters associated with the noise model, to each system equivalent pattern X times to produce, for each output class, X training samples, each simulating defects expected in real-world classifier input patterns. 2. A method as set forth in claim 1, wherein the plurality of parameters associated with the noise model are selected randomly. 3. A method as set forth in claim 1, wherein the system equivalent pattern for each class is a scanned image. 4. A method as set forth in claim 3, wherein the system equivalent pattern for each claus is a scanned image of an alphanumeric character. 5. A method as set forth in claim 3, wherein the system equivalent pattern for each class is the scanned image of a postage stamp. 6. A method as set forth in claim 3, wherein one of the plurality of known noise types simulates a horizontal stretching of the scanned image. 7. A method as set forth in claim 3, wherein one of the plurality of known noise types simulates Gaussian noise within the scanned image. 8. A method as set forth in claim 3, wherein one of the plurality of known noise types simulates a clipping of one side of the scanned image. 9. A computer program product, operative in a data processing system and implemented on a computer readable medium, for generating a set of X training images from a single ideal image for each output class of a pattern recognition classifier, comprising: a classifier system simulator that generates a system equivalent image for each of a plurality of classes from a corresponding ideal image; and a noise model, having a plurality of associated parameters, that simulates a plurality of noise types expected in a real-world classifier input image and incorporates the simulated noise into each system equivalent image X times to produce, for each output class, X training images, each training image simulating defects expected in real-world classifier input images. 10. A computer program product as set forth in claim 9, wherein plurality of parameters associated with the noise model is selected randomly. 11. A computer program product as set forth in claim 9, wherein the system equivalent image for each class is a scanned image of an alphanumeric character. 12. A computer program product as set forth in claim 9, wherein the system equivalent image for each class is the scanned image of a postage stamp. 13. A computer program product as set forth in claim 9, wherein one of the plurality of known noise types simulates a rotation of the scanned image. 14. A computer program product as set forth in claim 9, wherein one of the plurality of known noise types simulates distortion of the brightness of the scanned image. 15. A computer program product as set forth in claim 9, wherein one of the plurality of known noise types simulates a tear or cut within the source of the scanned image. 16. A method for training a pattern recognition classifier, comprising: generating a system equivalent pattern fur each of a plurality of output classes from a corresponding ideal pattern; generating a plurality of instances of a noise model for each of the system equivalent patterns, a given instance of the noise model simulating a plurality of noise types, where associated characteristics of each of the plurality of noise types are defined by a set of at least one parameter from a plurality of parameters comprising the noise model; applying each instance of the noise model to the system equivalent pattern to generate a plurality of training samples for each of the plurality of output classes; and training the pattern recognition classifier on the plurality of training samples generated for each of the plurality of output classes. 17. The method of claim 16, wherein the step of generating a plurality of instances of a noise model comprises, for a given instance, the step of setting the set of at least one parameter to associated null values for a selected set of at least one of the plurality of noise types such that the selected set of at least one noise type is not represented in the instance. 18. The method of claim 16, wherein the step of generating a plurality of instances of a noise model comprises, for a given instance, the step of randomly generating the at least one parameter associated with at least one of the plurality of noise types. 19. The method of claim 16, wherein the step of generating a plurality of instances of a noise model comprises, for a given instance, the step of accepting the sets of at least one parameter associated with the plurality of noise types as submitted by a human operator. 20. The method of claim 19, wherein the step of generating a plurality of instances of a noise mode comprises, for a given instance, the step of selecting the sets of at least one parameter associated with the plurality of noise types to produce an interaction between at least two of the plurality of noise types, such that the interaction between the noise types produces within the training sample associated with the instance a noise type that is not explicitly included in the plurality of noise types comprising the noise model.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (24)
Weideman William E. (Arlington TX), Adaptive neural network image processing system.
Vachtesvanos, George J.; Dorrity, Lewis J.; Wang, Peng; Echauz, Javier; Mufti, Muid, Method and apparatus for analyzing an image to detect and identify patterns.
Tong David W. (Scotia NY) Delano Paul A. (Schenectady NY), Method for using a feed forward neural network to perform classification with highly biased data.
Obata Kenzo (Okazaki JPX) Uchikawa Yoshiki (Nagoya JPX) Furuhashi Takeshi (Nagoya JPX) Watanabe Shigeru (Nagoya JPX), Signature recognition apparatus which can be trained with a reduced amount of sample data.
Luo, Yun; Wallace, Jon K., Method and apparatus for producing classifier training images via construction and manipulation of a three-dimensional image model.
Alcocer Ochoa, Alberto; Tinsley, Keith Raynard, Systems, methods, apparatus and computer readable mediums for use in association with systems having interference.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.