WO2019047025A1 - Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation - Google Patents
Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation Download PDFInfo
- Publication number
- WO2019047025A1 WO2019047025A1 PCT/CN2017/100581 CN2017100581W WO2019047025A1 WO 2019047025 A1 WO2019047025 A1 WO 2019047025A1 CN 2017100581 W CN2017100581 W CN 2017100581W WO 2019047025 A1 WO2019047025 A1 WO 2019047025A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- vector
- neighborhood
- length
- feature
- descriptor
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
Definitions
- the present invention relates to the field of image processing technologies, and in particular, to a method and a system for forming a local feature descriptor of a hyperspectral image.
- This method of combining spatial information and spectral information is generally applied to face recognition.
- Pan et al. pioneered a new method of hyperspectral face recognition by extracting spectral features on features such as forehead, cheeks, and lips, but this method did not apply to any spatial information.
- Unzir et al. developed an aerial spectrum special authentication extraction method to calculate the low frequency factor of hyperspectral face images.
- Hyperspectral imaging offers new opportunities for interpersonal facial recognition, however, the extraction of compact and discriminative features from high-dimensional hyperspectral image cubes is a challenging task.
- the 3D discrete cosine transform optimally compresses the information in the low frequency coefficients, represents each hyperspectral facial cube with a small number of low frequency DCT coefficients, and develops a partial least squares (PLS) regression for accurate classification.
- PLS partial least squares
- a spatial spectral feature descriptor is then generated by applying a 3D histogram on the derivative pattern, which can be used to convert the hyperspectral facial image to a vectorized representation.
- this method can describe unique microscopic patterns that integrate potential spatial and spectral information.
- Hyperspectral images and ordinary two-dimensional images have many different characteristics. They add spectral dimensions based on two-dimensional image information to form a three-dimensional coordinate space. If each band data of the hyperspectral image is regarded as a layer, and the imaging spectrum data is expressed as a whole in the coordinate space, a three-dimensional data cube having multiple layers formed by overlapping in a band sequence is formed. Hyperspectral images have many bands, which can provide hundreds or even thousands of bands per pixel. These bands are generally less than 10 cm in length, and the bands are continuous. Some sensors can provide near-solar spectrum. Continuous ground spectrum.
- hyperspectral data and the information it carries are generally represented by three spatial representations: image space with spatial geometric positional relationship, spectral space containing rich spectral information, and suitable for patterns.
- image space with spatial geometric positional relationship image space with spatial geometric positional relationship
- spectral space containing rich spectral information spectral space containing rich spectral information
- suitable for patterns The feature space of the application in recognition.
- Hyperspectral remote sensing techniques can combine spectra that determine the properties of a substance or feature with images that reveal their spatial and geometric relationships. The corresponding spectral curves of different materials are also different.
- an object of the present invention is to provide a method and a system for forming a local feature descriptor of a hyperspectral image, which aims to solve the problem that the local feature description algorithm of the two-dimensional image in the prior art is not applicable to the hyperspectral image. .
- the invention provides a method for forming a local feature descriptor of a hyperspectral image, wherein the method comprises:
- Feature description step using the angle between the spectral vectors and the modulus of the spectral vector to describe the neighborhood features;
- Forming a description sub-step forming a descriptor whose vector length is three times the length of the original local binary pattern vector according to the description result.
- the step of describing the feature specifically includes:
- the angle between the center value of the intermediate vector vector c and the synthesized vector c is a vector of the center-dimensional vector, wherein the vector length of the central vector c is N, the center of the die is a vector c vector norm c, N is an odd number, half of The median of the length N;
- the angle between the center vector vector c and the first dimension vector of the center vector vector c is synthesized.
- the step of describing the feature specifically includes:
- the angle between the neighborhood vector vector p and the first dimension vector of the neighborhood vector vector p is synthesized.
- the step of describing the feature specifically includes:
- the first value describing the neighborhood spatial feature is obtained from the original LBP operator expression. among them,
- the second value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the third value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the forming the sub-step includes:
- the three descriptors Descriptor norm , A join is made to obtain a new descriptor of length 3*2 ⁇ P, the length of the new descriptor being three times the length of the original local binary pattern vector.
- the present invention also provides a system for forming a local feature descriptor of a hyperspectral image, the system comprising:
- a feature description module for describing a neighborhood feature by using an angle between spectral vectors and a spectral vector
- a description sub-module is formed for forming a descriptor whose vector length is three times the length of the original local binary pattern vector according to the description result.
- the feature description module is specifically configured to:
- the angle between the center value of the intermediate vector vector c and the synthesized vector c is a vector of the center-dimensional vector, wherein the vector length of the central vector c is N, the center of the die is a vector c vector norm c, N is an odd number, half of The median of the length N;
- the angle between the center vector vector c and the first dimension vector of the center vector vector c is synthesized.
- the feature description module is further used to:
- Is the angle between vector values of the intermediate vector p neighborhood of the center vector of the synthesis of one-dimensional vector c is a vector, wherein the vector center vector c local neighborhood of V, neighborhood radius is R, the neighborhood points is P, o
- the angle between the neighborhood vector vector p and the first dimension vector of the neighborhood vector vector p is synthesized.
- the feature description module is further used to:
- the first value describing the neighborhood spatial feature is obtained from the original LBP operator expression. among them,
- the second value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the third value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the forming description submodule is specifically configured to:
- the three descriptors Descriptor norm , A join is made to obtain a new descriptor of length 3*2 ⁇ P, the length of the new descriptor being three times the length of the original local binary pattern vector.
- the technical solution provided by the invention utilizes the angle between the spectral vectors and the characterization of the neighborhood features of the spectral vector to form a vector whose length is three times the length of the original local binary pattern vector.
- the narration combined with the spatial position information and spectral vector information of the neighborhood of hyperspectral feature points, can be applied to hyperspectral images.
- FIG. 1 is a flow chart showing a method for forming a local feature descriptor of a hyperspectral image according to an embodiment of the present invention
- FIG. 2 is a neighborhood representation of a central spectral vector in accordance with an embodiment of the present invention.
- FIG. 3 is a schematic diagram showing the internal structure of a forming system 10 for a local feature descriptor of a hyperspectral image according to an embodiment of the present invention.
- a specific embodiment of the present invention provides a method for forming a local feature descriptor of a hyperspectral image, wherein the method mainly includes the following steps:
- Feature description step using the angle between the spectral vectors and the modulus of the spectral vector to describe the neighborhood features;
- Forming a description sub-step forming a descriptor whose vector length is three times the length of the original local binary pattern vector according to the description result.
- the invention provides a method for forming a local feature descriptor of a hyperspectral image, which uses the angle between the spectral vectors and the model of the spectral vector to describe the feature of the neighborhood feature to form a vector length which is the original local binary mode (Local Binary Patterns (LBP) triple the descriptor of the vector length, combined with the spatial position information and spectral vector information of the hyperspectral feature point neighborhood, can be applied to hyperspectral images.
- LBP Local Binary Patterns
- FIG. 1 is a flowchart of a method for forming a local feature descriptor of a hyperspectral image according to an embodiment of the present invention.
- step S1 the feature description step describes the neighborhood feature by using the angle between the spectral vectors and the mode of the spectral vector.
- the feature description step specifically includes:
- the angle between the center value of the intermediate vector vector c and the synthesized vector c is a vector of the center-dimensional vector, wherein the vector length of the central vector c is N, the center of the die is a vector c vector norm c, N is an odd number, half of The median of the length N;
- the angle between the center vector vector c and the first dimension vector of the center vector vector c is synthesized.
- the hyperspectral image contains another aspect of information, that is, the spectral response of the object, since the spectral vector of the hyperspectral image contains rich feature information. Therefore, it is more efficient to replace the gray value with the spectral vector.
- FIG. 2 draws a neighborhood representation of the central spectral vector, where the line indicated by 201 is the central spectral vector.
- FIG. 2 there is shown a neighborhood representation of a central spectral vector in accordance with an embodiment of the present invention.
- the center of the cube and the arrow to the right represent the central spectral vector
- a feature point in the spectral domain of the hyperspectral image is selected as the central pixel
- the DN value of the central pixel x c , y c , ⁇ c
- the upper DN value constitutes a set of vectors v c
- the DN value of the neighborhood cell of the local neighborhood can be composed of a series of vectors v p .
- the feature description step specifically includes:
- the length of the neighborhood vector vector p is the same as the center vector vector c ;
- the feature description step specifically includes:
- the first value describing the neighborhood spatial feature is obtained from the original LBP operator expression. among them,
- the second value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the third value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- step S2 a description sub-step is formed: a descriptor whose vector length is three times the length of the original local binary pattern vector is formed according to the description result.
- the forming the sub-steps specifically includes:
- the three descriptors Descriptor norm , A join is made to obtain a new descriptor of length 3*2 ⁇ P, the length of the new descriptor being three times the length of the original local binary pattern vector.
- the hyperspectral image organically combines the spectral information reflecting the radiation property of the substance with the two-dimensional image information reflecting the geometric relationship of the object space, so that the hyperspectral image can be grayscaled.
- the image and the color image provide more information.
- the image of the "integration of the map" proposed by the present invention combines the advantages of the two-dimensional image and the spectral information, and broadens the analysis method of the image, which is very important for image analysis and recognition. The meaning.
- the invention provides a method for forming a local feature descriptor of a hyperspectral image, and proposes a novel descriptor based on a local binary pattern for the feature of the spectral vector of the hyperspectral image.
- the angle between the spectral vectors and the model of the spectral vector are used to describe the features of the neighborhood, and a descriptor whose vector length is three times the length of the original local binary pattern vector is formed.
- the algorithm takes into account the characteristics of the hyperspectral image and combines the spatial position information and spectral vector information of the neighborhood of the hyperspectral feature points.
- a specific embodiment of the present invention further provides a system for forming a local feature descriptor of a hyperspectral image, which mainly includes:
- a feature description module for describing a neighborhood feature by using an angle between spectral vectors and a spectral vector
- a description sub-module is formed for forming a descriptor whose vector length is three times the length of the original local binary pattern vector according to the description result.
- the present invention provides a hyperspectral image local feature descriptor forming system 10, which utilizes the angle between the spectral vectors and the spectral vector to model the feature description of the neighborhood feature to form a vector length which is the original local binary pattern vector.
- the descriptor of three times the length, combined with the spatial position information and spectral vector information of the neighborhood of the hyperspectral feature points, can be applied to hyperspectral images.
- FIG. 3 a schematic structural diagram of a system 10 for forming a local feature descriptor of a hyperspectral image according to an embodiment of the present invention is shown.
- the hyperspectral image local feature descriptor forming system 10 mainly includes a feature description module 11 and a formation description sub-module 12.
- the feature description module 11 is configured to describe the neighborhood features by using the angle between the spectral vectors and the modulus of the spectral vector.
- the feature description module 11 is specifically configured to:
- the angle between the center value of the intermediate vector vector c and the synthesized vector c is a vector of the center-dimensional vector, wherein the vector length of the central vector c is N, the center of the die is a vector c vector norm c, N is an odd number, half of The median of the length N;
- the angle between the center vector vector c and the first dimension vector of the center vector vector c is synthesized.
- the feature description module 11 is further specifically configured to:
- the length of the neighborhood vector vector p is the same as the center vector vector c ;
- the feature description module 11 is further specifically configured to:
- the first value describing the neighborhood spatial feature is obtained from the original LBP operator expression. among them,
- the second value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- the third value describing the neighborhood spatial feature is obtained from the original LBP operator expression.
- Forming a description sub-module 12 for forming a vector length according to the description result is an original local binary value A three-dimensional descriptor of the length of the pattern vector.
- the forming description sub-module 12 is specifically configured to:
- the three descriptors Descriptor norm , A join is made to obtain a new descriptor of length 3*2 ⁇ P, the length of the new descriptor being three times the length of the original local binary pattern vector.
- the present invention provides a hyperspectral image local feature descriptor forming system 10, which utilizes the angle between the spectral vectors and the spectral vector to model the feature description of the neighborhood feature to form a vector length which is the original local binary pattern vector.
- the descriptor of three times the length, combined with the spatial position information and spectral vector information of the neighborhood of the hyperspectral feature points, can be applied to hyperspectral images.
- each unit included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding function can be implemented; in addition, the specific name of each functional unit is also They are only used to facilitate mutual differentiation and are not intended to limit the scope of the present invention.
Abstract
L'invention concerne un procédé de formation d'un descripteur de caractéristique locale d'une image hyperspectrale, le procédé comportant: une étape de description de caractéristique consistant à: décrire une caractéristique de voisinage en utilisant un angle inclus entre des vecteurs spectraux et une norme des vecteurs spectraux (S1); une étape de formation de descripteur consistant: d'après un résultat de description, à former un descripteur présentant une longueur de vecteur qui vaut trois fois la longueur d'un vecteur en mode binaire local d'origine (S2). L'invention concerne en outre un système de formation d'un descripteur de caractéristique locale d'une image hyperspectrale, qui: forme un descripteur présentant une longueur de vecteur qui vaut trois fois la longueur d'un vecteur en mode binaire local d'origine en utilisant la capacité de description de caractéristique d'un angle inclus entre des vecteurs de spectre et une norme des vecteurs spectraux pour une caractéristique de voisinage, tout en combinant des informations de position spatiale et des informations spectrales d'un voisinage d'un point caractéristique hyperspectral. Le système peut être appliqué à des images hyperspectrales.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2017/100581 WO2019047025A1 (fr) | 2017-09-05 | 2017-09-05 | Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2017/100581 WO2019047025A1 (fr) | 2017-09-05 | 2017-09-05 | Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019047025A1 true WO2019047025A1 (fr) | 2019-03-14 |
Family
ID=65633452
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2017/100581 WO2019047025A1 (fr) | 2017-09-05 | 2017-09-05 | Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2019047025A1 (fr) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110311142A1 (en) * | 2010-06-18 | 2011-12-22 | National Ict Australia Limited | Descriptor of a hyperspectral or multispectral image |
CN105608433A (zh) * | 2015-12-23 | 2016-05-25 | 北京化工大学 | 一种基于核协同表达的高光谱图像分类方法 |
CN106485238A (zh) * | 2016-11-01 | 2017-03-08 | 深圳大学 | 一种高光谱遥感图像特征提取和分类方法及其系统 |
CN106529508A (zh) * | 2016-12-07 | 2017-03-22 | 西安电子科技大学 | 基于局部和非局部多特征语义高光谱图像分类方法 |
-
2017
- 2017-09-05 WO PCT/CN2017/100581 patent/WO2019047025A1/fr active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110311142A1 (en) * | 2010-06-18 | 2011-12-22 | National Ict Australia Limited | Descriptor of a hyperspectral or multispectral image |
CN105608433A (zh) * | 2015-12-23 | 2016-05-25 | 北京化工大学 | 一种基于核协同表达的高光谱图像分类方法 |
CN106485238A (zh) * | 2016-11-01 | 2017-03-08 | 深圳大学 | 一种高光谱遥感图像特征提取和分类方法及其系统 |
CN106529508A (zh) * | 2016-12-07 | 2017-03-22 | 西安电子科技大学 | 基于局部和非局部多特征语义高光谱图像分类方法 |
Non-Patent Citations (2)
Title |
---|
LI, WEI.: "Local Binary Patterns and Extreme Learning Machine for Hyperspectral Imagery Classification", IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, vol. 53, no. 7, 31 July 2015 (2015-07-31), XP011575081 * |
YE, ZHEN ET AL.: "Hyperspectral Image Classification Based on Principal Component Analysis and Local Binary Patterns", LASER & OPTOELECTRONICS, no. 11, 6 July 2017 (2017-07-06), pages 111006-1 - 111006-9 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110443842B (zh) | 基于视角融合的深度图预测方法 | |
Xie et al. | Joint super resolution and denoising from a single depth image | |
Wang et al. | Neuris: Neural reconstruction of indoor scenes using normal priors | |
Faridul et al. | Colour mapping: A review of recent methods, extensions and applications | |
US7817161B2 (en) | Texture synthesis using dimensionality-reduced appearance space | |
CN108876723B (zh) | 一种灰度目标图像的彩色背景的构建方法 | |
Xie et al. | Single depth image super resolution and denoising via coupled dictionary learning with local constraints and shock filtering | |
Pesaresi et al. | A new compact representation of morphological profiles: Report on first massive VHR image processing at the JRC | |
US11887341B2 (en) | Systems and methods for hue based encoding of a digital image | |
Li et al. | SiamBAG: Band attention grouping-based Siamese object tracking network for hyperspectral videos | |
Chaki et al. | Image color feature extraction techniques: fundamentals and applications | |
Yaman et al. | An iterative adaptive multi-modal stereo-vision method using mutual information | |
Guo et al. | Edge-guided hyperspectral image compression with interactive dual attention | |
US8520980B2 (en) | Image searching method and apparatus | |
IlaVennila | Optimization fusion approach for image segmentation using k-means algorithm | |
ALEjaily et al. | Fusion of remote sensing images using contourlet transform | |
Li et al. | I-gans for infrared image generation | |
WO2019047025A1 (fr) | Procédé de formation de descripteur de caractéristique locale d'image hyperspectrale et système de formation | |
Obulesu et al. | Image retrieval based local motif patterns code | |
Wang et al. | Multi-sensor Image Fusion Algorithm Based on Multiresolution Analysis. | |
Dinesh et al. | Point cloud video super-resolution via partial point coupling and graph smoothness | |
CN107679538B (zh) | 高光谱图像局部特征描述子的形成方法及形成系统 | |
Guerrero et al. | DepthCut: improved depth edge estimation using multiple unreliable channels | |
Dixit et al. | Copy-move image forgery detection using frequency-based techniques: a review | |
Zhang et al. | Graph-Based Depth Denoising & Dequantization for Point Cloud Enhancement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17924465 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 30/09/2020) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 17924465 Country of ref document: EP Kind code of ref document: A1 |