Method for spectral data classification and detection in diverse lighting conditions
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06K-009/40
G06K-009/46
G06K-009/62
G06K-009/00
출원번호
UP-0676225
(2007-02-16)
등록번호
US-7796833
(2010-10-04)
발명자
/ 주소
Polonskiy, Leonid
Wang, Zhu Joe
Benac, Jasenka
Golden, Jeffry
출원인 / 주소
CET, LLC
대리인 / 주소
Kang, Grant D.
인용정보
피인용 횟수 :
8인용 특허 :
18
초록▼
The invention is a method of spectral data classification that uses the decoupling of target chromaticity and lighting or illumination chromaticity in spectral data and the sorting and selection of spectral bands by values of a merit function to obtain an optimized set of combinations of spectral ba
The invention is a method of spectral data classification that uses the decoupling of target chromaticity and lighting or illumination chromaticity in spectral data and the sorting and selection of spectral bands by values of a merit function to obtain an optimized set of combinations of spectral bands for classification of the data. The decoupling is performed in “delta-log” space. A rotation transform may be applied. For a broad range of parameters, correction of lighting chromaticity may be obtained by use of an equivalent “Planck distribution” temperature. Merit function sorting and band combination selection is performed by multiple selection criteria. The method achieves reliable pixel classification and target detection in diverse lighting or illumination, especially in circumstances where lighting is non-uniform across a scene, such as with sunlight and shadows on a partly cloudy day or in “artificial” lighting. Applications are found in homeland security, defense, environmental protection, biomedical diagnostics, industrial process and product monitoring, and other remote or standoff sensing by spectral characteristics.
대표청구항▼
What is claimed is: 1. A method for hyper-spectral or multi-spectral data classification or target detection in diverse lighting conditions with an invariant set of detection parameters, based on decoupling target chromaticity and lighting chromaticity in a delta-logarithmic feature space and compr
What is claimed is: 1. A method for hyper-spectral or multi-spectral data classification or target detection in diverse lighting conditions with an invariant set of detection parameters, based on decoupling target chromaticity and lighting chromaticity in a delta-logarithmic feature space and comprising the steps of: acquiring two or more multi-spectral or hyperspectral image data, each set being an image hyper-cube for a scene and the two sets having been acquired with illumination having different chromaticity; performing a logarithmic transform of the acquired image hyper-cubes; calculating chromaticity maps for all triplets or quadruplets of wavelengths, each said triplet or quadruplet comprising a “band combination”, and each element of the map corresponding to a pixel, or group of pixels treated as a single element, with the map coordinates each being the logarithm of the ratio of the pixel values at two of the different wavelengths of the triplet or quadruplet, the logarithm of the ratio of the pixel values being the spectral band subtraction in the logarithmic feature space, then determining isochrome lines for each band combination; sorting all band combinations by merit function values, the value of each band combination being calculated as a function of the value of at least one criteria that are applied to the isochromes, and the sorting being the ranking of the band combinations according to their merit function value, the greatest rank being given to the most advantageous merit function value; comparing presumed-target chromaticities in a subset of feature space that corresponds to one or more of the highest ranked band combinations with classification or detection discriminant parameters; and generating a detection mask through the denotation of the subset of pixels that have been identified or classified, as having a chromaticity that sufficiently matches the detection discriminant parameters. 2. The method of claim 1 in which the sorting by merit function values is according to Tri-Criteria or Multi-Criteria evaluation. 3. The method of claim 2 in which the Tri-Criteria or Multi-Criteria are selected from a group consisting of one or more of the following: a function of the minimum value of the mean error of linear fitting the isochromes, a function of the minimum difference in the angles of tilt (slopes) of the isochromes, a function of the ratio of the distance between adjacent isochromes and the sum of the mean dispersion of the corresponding classes. 4. The method of claim 1 in which the resulting isochromes are rotated to a vertical orientation. 5. The method of claim 1 with the additional step of: registering or outputting the results of detection, for example exhibiting the results as a display, or storing the results in a memory, or generating a signal such as an alarm. 6. A method for hyper-spectral or multi-spectral data classification and target signature determination, based on decoupling target chromaticity and lighting chromaticity in a delta-logarithmic feature space and comprising the steps of: acquiring a plurality of sets of image hyper-cube data of at least one scene, each of said scenes containing the target or material of interest, the data sets differing by their illumination, which has different light chromaticity for the different data sets; performing a logarithmic transform of the said acquired image hyper-cubes, resulting in hyper-cubes with each element being the logarithm of the corresponding pixel value; calculating chromaticity maps for all triplets or quadruplets of wavelengths, each said triplet or quadruplet comprising a “band combination”, and each element of the map corresponding to a pixel, or group of pixels, i.e., a region of interest, treated as a single element, with the map coordinates each being the logarithm of the ratio of the pixel values at two of the different wavelengths of the triplet or quadruplet, the logarithm of the ratio of the pixel values being the spectral band subtraction in the logarithmic feature space, then determining isochrome lines for each band combination; sorting all band combinations by merit function values, the value of each band combination being calculated as a function of the value of at least one criteria that are applied to the isochromes, and the sorting being the ordering of the band combinations according to their merit function value, the greatest rank being given to the most advantageous merit function value; and selecting a set containing one or more of the highest ranked band combinations, this set corresponding to a set of projection planes of a subspace of feature space, and identifying one or more isochromes in the subspace of feature space, and the identified isochromes and band combinations comprising the target signature. 7. The method of claim 6 wherein the target data comprise hyper-spectral images of a color chart. 8. The method of claim 6 in which the resulting isochromes are rotated to a vertical orientation. 9. The method of claim 6 in which the sorting criteria are selected from a group comprising one or more of the following: a function of the minimum value of the mean error of linear fitting the isochromes, a function of the minimum difference in the angles of tilt (slopes) of the isochromes, a function of the ratio of the distance between adjacent isochromes and the sum of the mean dispersion of the corresponding classes. 10. A method for hyper-spectral or multi-spectral data classification or target detection in diverse lighting conditions with an invariant set of classification parameters, based on decoupling target chromaticity and lighting chromaticity in a delta-logarithmic feature space and comprising the steps of: acquisition of image hyper-cube data and data correction as described above for a known target, material, object, or color calibration chart in various lighting conditions that correspond to various color temperatures; performing a logarithmic transform of the acquired image hyper-cube data, the result is a hyper-cube with each element being the logarithm of the corresponding pixel value; calculating chromaticity maps for all triplets or quadruplets of wavelengths, each said triplet or quadruplet comprising a “band combination”, and each element of the map corresponding to a pixel, or group of pixels treated as a single element, with the map coordinates each being the logarithm of the ratio of the pixel values at two of the different wavelengths of the triplet or quadruplet, the logarithm of the ratio of the pixel values being the spectral band subtraction in the logarithmic feature space according to the equation ln P k l = ln [ P k P l ] = ln P k - ln P l = R k - R l - L k - L l T , and, the resulting isochromes being rotated to a vertical orientation where selected; sorting all band combinations by merit function values, the value of each band combination being calculated as a function of the value set of at least one criteria that are applied to the isochromes, and the sorting being the ordering of the band combinations according to their merit function value, the greatest rank being given to the most advantageous merit function value; selection of a subset comprising one or more of the highest ranked band combinations, which define a feature subspace, and storage of these bands in a feature subspace band combination list and also storage of classification or detection discriminant parameters comprising the target isochromes and their spacing from neighboring non-target isochromes in the feature subspace of these band combinations; acquisition of image hyper-cube data and data correction for a scene that is to be searched for targets, specific materials, or specific objects, this scene being referred to as a “search scene”; performing a logarithmic transform of the acquired image hyper-cube data of the search scene; calculating chromaticity maps for the feature subspace, i.e., for the subset of band combinations selected above, and each element of the map corresponding to a pixel, or group of pixels treated as a single element, with the map coordinates each being the logarithm of the ratio of the pixel values at two of the different wavelengths of the triplet or quadruplet, the logarithm of the ratio of the pixel values being the spectral band subtraction in the logarithmic feature space, and where a rotational transform was applied to the isochrome map, then, applying the same rotational transform for that band combination; comparing presumed-target chromaticities in the feature subspace with the above identified classification or detection discriminant parameters that represent a known target, material or object. 11. The method of claim 10 with the additional step of: generating a detection mask through the denotation of the subset of pixels that have been identified as having a chromaticity that sufficiently matches the classification or detection discriminant parameters. 12. The method of claim 1 wherein the lighting chromaticity is measured as a function of equivalent color temperature that is based on a Planckian or near-Planckian distribution. 13. The method of claim 1 wherein the spectral range of measurement is in the short wavelength approximation, which is defined as exp ( c 2 T λ ) >> 1 , and where c2=1.44·10−2 K·m, wherein T is the temperature of the black body radiator in K, and λ is the wavelength in m. 14. The method of claim 1 wherein the spectral range of measurement is in the long wavelength approximation, which is defined as Tλ<<1. 15. The use of the method of claim 1 for determining the reflectance of an object in a scene by analysis of hyperspectral or multi-spectral image data for a plurality of conditions of illumination with different color temperatures. 16. The use of the method of claim 1 for determining the thermal emissivity of an object in a scene.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (18)
Usui Shiro (Takashijyutaku #5-504 2-1 ; Higashiura ; Kitayama-chou ; Toyohasi ; Aichi ; 441 JPX), Color balance adjusting apparatus using a decorrelating neural network.
Funt Brian V. (West Vancouver CAX) Ho Jian (Nepean CAX) Drew Mark S. (Vancouver CAX), Method and apparatus for determining ambient light and surface reflectance.
Thornton ; Jr. William A. (27 Harvard Rd. Cranford NJ 07016), Method and apparatus for measuring any of a large number of characteristics of lamplight.
Bernstein, Lawrence S.; Adler-Golden, Steven M.; Perkins, Timothy C.; Berk, Alexander; Levine, Robert Y., Method for performing automated in-scene based atmospheric compensation for multi-and hyperspectral imaging sensors in the solar reflective spectral region.
Asano Yuichiro (Palo Alto CA) Shiozumi Motoji (Kobe JPX) Aizawa Hitoshi (Kurashiki JPX) Shibata Yoshiaki (Chiba JPX), Method of and apparatus for determining glossiness of surface of a body.
Abreu Leonard W. (Chelmsford MA) Kneizys Francis X. (Burlington MA) Anderson Gail P. (Concord MA) Chetwynd James H. (Stoneham MA) Berk Lex (Medford MA) Bernstein Larry (Lexington MA) Robertson David , System for modelling moderate resolution atmospheric propagation.
Lynass, Mark Ronald; Heller, Christian; Schmidt, Michael, Apparatus having a controllable filter matrix to selectively acquire different components of solar irradiance.
Schwartz, Abraham; Sherman, Philip; Ingram, Peter, Method to remove the spectral components of illumination and background from multi-spectral and hyper-spectral images.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.