A method for extracting depth from image information

Abstract

The present invention proposes a method for deep extraction of image information, characterized by utilizing multiple image processing methods to perform multiple image processing operations on at least one sample image; obtaining multiple expanded images that can simulate the sample image in different environments; using the multiple expanded images as a set of sample image data for the image; and generating a probability distribution of sample feature values ​​for a single image. The advantages and positive effects of the present invention include: through image processing, a single sample face image can be transformed into multiple images, forming a probability distribution of sample images, thereby improving the recognition accuracy of a single face image.

Description

Technical field

[0001] The invention belongs to a method for extracting depth of image information in the field of image processing.

[0002] Currently, technological competition in the field of facial recognition has become a major battlefield in artificial intelligence, and the level of competition has reached an extremely fierce level. NEC Corporation has published a patent entitled "Pattern Recognition Method" (Patent Document 1). This patent discloses a method that uses a feature vector composed of 30 eigenvalues ​​of the facial features to train the covariance between sets of facial feature vectors for the same person and between sets of different facial feature vectors. Finally, using the Mahalanobis distance formula, the facial data corresponding to the feature vector with the minimum distance between the sample feature vector and different registered facial feature vectors is set as the recognition result.

[0003] NEC Corporation, which possesses the highest level of facial recognition technology in Japan, subsequently disclosed: "Method and device for generating facial feature data, and method and device for calculating facial similarity" (Patent Document 2). This patent further adds a reliability index to the recognition of each facial image based on the previous patent. This reliability index is derived from the normal distribution of the vector difference between feature vectors, the variance between the eigenvalues ​​of the difference vector of the feature vector, the posterior distribution of the difference vector between facial features as a mixed distribution, and the logarithmic degree of the mixed distribution. The similarity between patterns is calculated according to the reliability, thereby improving the accuracy of facial recognition.

[0004] In 2004, NEC and Samsung Electronics of South Korea also disclosed a patent for face recognition: "Face recognition and feature vector extraction method and device for retrieval" (Patent Document 3). This patent discloses the first normalized vector and the second normalized vector for the entire face image, which are used to generate the overall Fourier transform feature vector through Fourier transform, and then the third normalized vector and the fourth normalized vector for the central part of the face image are used to generate the Fourier transform feature vector of the central part through Fourier transform. An overall intensity feature vector is generated for the overall face image, and a local intensity feature vector is generated for the face image of a specified local area. The above-mentioned first normalized vector and the second normalized vector are combined with the overall intensity feature vector to generate an overall feature vector, and the above-mentioned third normalized vector and the fourth normalized vector are combined with the local intensity feature vector to generate a local feature vector. This constitutes the feature vector used for retrieval.

[0005] In response to the above invention, NEC Corporation published a patent in 2006 entitled "Method and device for generating facial feature data, and method and system for facial recognition" (Patent Document 4). Based on Patent Document 3, this patent proposes a new calculation method for reflecting the reliability of the original image state, such as generating reliability values ​​based on differences in image contrast, generating reliability values ​​based on differences in the maximum and minimum grayscale values ​​of the image pixels, generating reliability values ​​based on differences in the pixel dispersion and standard deviation of the image, generating reliability values ​​based on differences in the image's asymmetry index, and generating reliability values ​​based on differences in the values ​​obtained by multiplying or adding the pixel values ​​of the differential image between the extracted image and the inverted image.

[0006] Over a decade later, NEC Corporation released its latest patent for facial recognition technology: "Face Recognition Device, Face Recognition Method, Face Recognition Program, Display Control Device, Display Control Method, and Display Control Program" (Patent Document 5). This patent proposes a method for identifying facial images based on their locality. Based on a threshold value for the similarity of these local facial images, three results are derived: similarity, ambiguity, and dissimilarity. These three conclusions for each local facial image serve as the basis for overall facial image recognition.

[0007] [Public Documents]

[0008] [Patent Document 1] Japanese Patent Application Publication No. 9-134432

[0009] [Patent Document 2] Japanese Patent Application Publication No. 2003-187229

[0010] [Patent Document 3] Special Opening No. 2004-139596 (P2004-139596A)

[0011] [Patent Document 4] Japanese Patent Application Laid-Open No. 2006-344236 (P2006-344236A)

[0012] [Patent Document 5] Japanese Patent Application Laid-Open No. 2018-49655 (P2018-49655A)

[0013] The method described in Patent Document 1 above trains the covariance between sets of facial feature vectors for the same person and between sets of different facial feature vectors, and uses the Mahalanobis distance to obtain the final recognition result. While this method can apply probabilistic statistical formulas to face recognition, significantly improving upon traditional image recognition methods, the presence of significant random interference in face recognition images makes it difficult to transcend probabilistic statistical formulas, filter out noise in facial images and feature values, and achieve predicted recognition results close to the original. Furthermore, the Mahalanobis distance only considers similar relationships between feature vectors and cannot directly use distances, which unify Euclidean space and probability space, to accurately calculate the scale relationships between different feature vectors, hindering recognition accuracy.

[0014] The method described in the above-mentioned patent document 2 adds a reliability index processing for the recognition of each face image. This method can undoubtedly improve the accuracy of face recognition, but the definition of reliability lacks strict scientificity and cannot achieve optimal recognition results.

[0015] The method described in the above-mentioned patent document 3 proposes to use the spectrum of Fourier transform as the feature vector of the face image. Since the rotation of the face image has a great influence on the spectrum, and the spectrum information is not very sensitive to the feature information of the face image, and the feature values ​​are filled with a large amount of noise, it seriously affects the improvement of the recognition rate.

[0016] The method described in Patent Document 4 mentioned above increases the accuracy of face recognition by generating a reliability that can reflect the state of the original image. However, the generation of the reliability is not rigorous enough.

[0017] The method described in the above-mentioned patent document 5 is a method of identifying each part of the face image one by one. The final reliability processing uses three reliability thresholds as the judgment basis, which lacks rigor and still has the above-mentioned problems in the recognition of partial face images. [Summary of the invention]

[0018] The first purpose of the present invention is to provide an information extraction method for image recognition that is suitable for generating large-scale feature vectors of the order of billions, which can solve the problem of misalignment of the recognized image and can most effectively extract the feature information of the recognized image.

[0019] The second purpose of the present invention is to provide a method for solving the problem of how to obtain high-confidence feature value data from a large number of feature values ​​of identified images in large-scale image feature vectors.

[0020] The third purpose of the present invention is to propose a method for constructing a new generation of artificial intelligence models with big data feature vectors that can produce a large number of data sets that exceeds the number of traditional deep learning data sets through small data training.

[0021] The fourth object of the present invention is to propose a model for multi-purpose automatic machine learning of image processing and image recognition.

[0022] The fifth objective of the present invention is to propose a method for 3D object recognition based on automatic machine learning without the need for training.

[0023] The sixth objective of the present invention is to propose a processing method for radar data to achieve deep data mining and machine understanding.

[0024] The present invention solves the technical problem by adopting the following technical solutions:

[0025] A method for extracting depth of image information, characterized by:

[0026] A method for performing multiple image processing on at least one sample image using multiple image processing methods; obtaining multiple expanded images produced by different environments that can simulate the sample image; using the multiple expanded images as a set of sample data for the image; and enabling an image to also generate a probability distribution of sample feature values.

[0027] Furthermore, the method for enabling an image to generate a probability distribution of sample feature values ​​includes performing one of forward recognition, reverse recognition, or bidirectional recognition methods on the image.

[0028] Furthermore, the image processing refers to: using at least one method including image brightness transformation, color transformation, image enhancement, suppression of certain components, and geometric transformation to improve the visual quality of the image;

[0029] Alternatively, at least one of the following methods may be used: enhancing high-frequency components of an image to make facial features and details of a person in the image clear and distinct; and enhancing low-frequency components of an image to remove noise in the image;

[0030] Alternatively, the image may be divided into multiple images by image filtering, thereby improving the image quality and increasing the size of the feature vector.

[0031] Or at least one method based on a grayscale value transformation function including logarithmic transformation, gamma transformation, piecewise linear transformation, and histogram transformation of an image;

[0032] or at least one method based on spatial filtering including smoothing spatial filtering, statistical sorting filtering, sharpening spatial filtering, unsharpening concealment and high lifting filtering, and gradient sharpening;

[0033] or at least one frequency-based filtering method selected from the group consisting of ideal low-pass filtering, ideal high-pass filtering, Butel low-pass filtering, Butel high-pass filtering, Gaussian low-pass filtering, Gaussian high-pass filtering, passivation mask, high-boost filtering, high-frequency emphasis filtering, homomorphic filtering, selective filtering, and two-trap filtering;

[0034] or a method of at least one noise probability density function including Gaussian noise, Rayleigh noise, Erratic noise, exponential noise, uniform noise, and bibasic impulse noise;

[0035] Or at least one spatial filtering method for restoring noise including a mean filter, an arithmetic mean filter, a geometric mean filter, a harmonic mean filter, an inverse harmonic mean filter, a statistical sorting filter, an adaptive filter, and a frequency domain method for eliminating periodic noise;

[0036] Alternatively, the method may include at least one of an image processing method for emphasizing border information and an image processing method for adjusting grayscale values.

[0037] The advantages and positive effects of the present invention are:

[0038] Through image processing, a sample image of a face can be transformed into multiple images, and a probability distribution of the sample images can be formed, thereby improving the recognition accuracy of a single face image.

[0039] It can provide an accurate scaling formula that fully utilizes probabilistic information and fuzzy information, realize deep image recognition, realize quantitative evaluation of image recognition accuracy, and realize optimization of recognition results.

[0040] The following are definitions of the special terms that appear in this manual:

[0041] The so-called maximum probability: the result of machine learning processing that goes beyond traditional statistical formulaic algorithms and can be as close to the parent as possible.

[0042] The so-called probability scale self-organization: self-organizing iterative processing with maximum probability as the scale can obtain the maximum probability value and the scale of the maximum probability space.

[0043] The so-called multi-purpose automatic machine learning: the optimal solution of the objective function can be obtained without training for multiple purposes.

[0044] The so-called SDL (Super Deep Learning) model is an artificial intelligence model composed of a new type of neural network dominated by function probability distribution and using complex machine learning to connect nodes.

[0045] The so-called EPD distance: can unify the distance scale of Euclidean space and probability space.

[0046] The so-called FPM measure: a fuzzy event probability measure that can unify Euclidean space and probability space.

Brief Description of the Drawings

[0047] Figure 1 is the given image of the object to be identified

[0048] Figure 2 This is a schematic diagram of the amplitude spectrum image.

[0049] Figure 3 This is a diagram of a centralized data processing method.

[0050] Figure 4 Yes Figure 2 Diagram of diagonal data after data swapping

[0051] Figure 5 This is a schematic diagram of the transformation of the two-dimensional phase spectrum (Phase)

[0052] Figure 6 This is a schematic diagram of the SDL model that can handle large-scale feature vectors.

[0053] Figure 7 Grayscale value adjustment image processing method

[0054] Figure 8 It is an image processing method that emphasizes border information

[0055] Figure 9 It is another image processing method that emphasizes border information

[0056] Figure 10 A schematic diagram of a single photo image versus a single recognition image

[0057] Figure 11 This is a schematic diagram of the recognition of a single-shot image using simulated multi-scene photo images.

[0058] Figure 12 This is a schematic diagram of image recognition of a single photo in multiple scenes.

[0059] Figure 13 Schematic diagram of single photo image recognition for multi-scene shooting and simulated multi-scene shooting

[0060] Figure 14 It is a schematic diagram of dividing a probability space into multiple probability spaces

[0061] Figure 15 is a diagram of two probability spaces in Euclidean space

[0062] Figure 16 This is a schematic diagram of a pattern recognition method that uses common probability distribution information.

[0063] Figure 17 This is a diagram of a new probability value definition method that takes into account the overall probability.

[0064] Figure 18 This is a process flow chart for multi-purpose automatic machine learning.

[0065] Figure 19 This is a flow chart of the method for machine understanding radar data processing [Specific implementation method]

[0066] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings, but the embodiments of the present invention are illustrative rather than restrictive.

[0067] The present invention solves its technical problem by adopting the following technical solutions, and the embodiments of the invention are described with reference to the accompanying drawings.

[0068] First, a method for extracting depth from image information is proposed.

[0069] Figure 1 is the given image of the object to be identified.

[0070] against Figure 1 The image can be used to perform discrete Fourier transform (DFT) of a two-dimensional image using the following formula.

[0071] [Formula 1]

[0072]

[0073] Among them, f(x, y) is a digital image of size M*N, u and v are used to determine their frequencies, and the frequency domain system is the coordinate system generated by them.

[0074] The inverse discrete Fourier transform (IDFT) is given by:

[0075] [Formula 2]

[0076]

[0077] right Figure 1 The data after image transformation (original):

[0078]

Data 1

[0079] F(0,0), F(0,1),...F(0,N)

[0080] F(1,0),F(1,1),...F(1,N)

[0081] …

[0082] …

[0083] F(M,0), F(0,0),...F(M,N)

[0084] Amplitude spectrum calculation method (Magnitude)

[0085] Let R and I represent the real and imaginary parts of F respectively, then the amplitude spectrum formula is as follows:

[0086] [Formula 3]

[0087]

[0088] Since the frequency domain values ​​vary greatly and cannot be displayed directly, they are converted using the following formula 4, where white represents high values.

[0089] [Formula 4]

[0090] M1=log(1+M)

[0091] Figure 2 Schematic diagram of the amplitude spectrum image.

[0092] like Figure 2 shown; Figure 1 Applying formula 3 and formula 4 to image data, we can get Figure 2 The amplitude spectrum image shown, here, white represents high value experimental results.

[0093] Figure 3 This is a diagram of a centralized data processing method.

[0094] like Figure 3 shown; Figure 2 By swapping the diagonal data 1 and 3, and 2 and 4, the DC component can be moved to the center of the spectrum to obtain Figure 4 image.

[0095] Figure 5 It is a schematic diagram of the transformation of the two-dimensional phase spectrum (Phase).

[0096] like Figure 5 As shown; let R and I represent the real part and imaginary part of F respectively, then the phase spectrum formula is as follows:

[0097] Formula 5

[0098]

[0099] The specific meaning of the face value in the phase spectrum is radians, and the range is: -3.14159 to 3.14159.

[0100] Using the Fourier transform spectrum for face recognition is a well-known technology. The general Fourier transform spectrum will not be affected when the face image is translated. However, when the face image is rotated, it exceeds the discrete window of the Fourier transform, so the Fourier transform result will be erroneous. Therefore, it is necessary to process the face image so that it can still be contained in the discrete window even after rotation.

[0101] The above-mentioned eigenvalue extraction method is one of the methods of image transformation. In addition, as a method of considering image orthogonal transformation, you can also consider discrete cosine transform (DCT, Discrete Cosine Transform), Walsh transform (WT, Walsh, Transform), Hadamard transform (ST, Hadamard Transform), KL transform (KL, Karhunen-Loeve Transform) cotransform (ST, Slant Transform), discrete wavelet transform (DWT, Discrete Wavelet Transform), etc.

[0102] In addition, you can also consider principal component analysis (PCA, Principal Component Analysis), linear discriminant analysis (LDA, Linear Discriminant Analysis), periodic parameter estimation method (ARS proposed by Yukihiro Kamiya), and Gabor transform (GT, Gabor, Transform) and other methods.

[0103] Here, a method of extracting depth information at a specific position in an image is proposed.

[0104] The facial image is transformed into various images through a physical model. For example, a power image can be generated by substituting a power model into the facial image, and then the transformed facial image information is deeply extracted. Here, not only the position information of the facial features is extracted, but also each facial feature position should be described in more detail, such as the length, width, color, wrinkle information, and the ratio of the upper and lower lips of the mouth, etc. The size information of the eye, the size of the eyeball, and the color information of the eye are recorded for the position of the eye, the shape of the nose, the size of the nose, the color distribution of the nose, etc. The geometric features of each position of the face, such as the distribution of facial wrinkles, the skin color of the face, the local roughness, and the shape of the face, are all used as feature values ​​for deep extraction of image information.

[0105] Figure 6 It is a schematic diagram of the SDL model that can handle large-scale feature vectors.

[0106] like Figure 6 As shown in the figure, this is a new three-layer neural network consisting of a perception layer, a neural layer, and a cerebral cortex. It is an SDL model that can process large-scale feature vectors.

[0107] Here, a method for performing large-scale image transformation is proposed. The image can be segmented on a large scale, and each segmentation can be transformed, so that millions or even hundreds of millions of feature sets can be obtained to achieve high-precision image recognition results that surpass traditional deep learning.

[0108] like Figure 6 As shown in the figure, (FP1) is a part of the face image, such as the image of the eye. (FP2) is the image of other parts of the face image, such as the image of the nose. The face image can also be divided into n regions, (FP n ) is the last part of the face image, for example, the image of the mouth. The more detailed the facial image segmentation, the greater the accuracy of face recognition.

[0109] When segmenting a face image, the position information of the face image can also be used as the face positioning information for image segmentation, which can solve the problem that the accuracy of the recognition result is affected by the shift and rotation of the image.

[0110] (IT1), (IT2)~(IT m ) is a type of image transformation, including Fourier transform (DFT, Discrete Fourier Transform), discrete cosine transform (DCT, Discrete Cosine Transform), Walsh transform (WT, Walsh, Transform), Hadamard transform (ST, Hadamard Transform), KL transform (KL, Karhunen-Loeve Transform) cotransform (ST, Slant Transform), discrete wavelet transform (DWT, Discrete Wavelet Transform), principal component analysis (PCA, Principal Component Analysis), linear discriminant analysis (LDA, Linear Discriminant Analysis), periodic parameter estimation method (ARS proposed by Yukihiro Kamiya), and Gabor (GT, Gabor, Transform) transform methods.

[0111] like Figure 6As shown in the figure, (ML1') is a probability scale self-organizing machine learning unit, which is responsible for obtaining the feature value with the maximum probability from the multiple feature information and inputting it into the node (N1') of the perception layer. When the number of feature values ​​is odd, (ML1') can be omitted and the feature value can be directly input into the node (N1') of the perception layer.

[0112] In this SDL model that can process large-scale feature vectors, the number of nodes (N1') and (N2') in the perception layer (P1') and the neural layer (P2') can be increased to infinity (N1 (1) ')~(N1 (n) '), (N2 (1) ')~(N2 (n) '), where n→∞. Therefore, as the number of nodes increases, the probability scale self-organizing machine learning unit also becomes (ML1 (1) ')~(ML1 (n) '), (ML2 (1) ')~(ML2 (n) '), similarly n→∞.

[0113] After studying the mechanism of deep learning, it is found that the application effect of deep learning depends on the size of a huge data set. However, the cost of increasing the data set for deep learning is O(n n ), this computational complexity limits the ability of deep learning to expand the size of the dataset indefinitely. In contrast, the SDL model has a computational complexity of O(n 2 ) remains unchanged, so we can always expand the number of nodes (N1') and (N2') on the perception layer (P1') and the neural layer (P2') to achieve the small data training characteristics of the probability distribution of the function and the characteristics of obtaining large-scale feature vectors, thereby achieving application effects beyond deep learning.

[0114] Unlike the publicly available SDL model, the image to be recognized by the traditional SDL model is expanded into a mapping image through a physical model. This method makes it difficult to make the feature vector reach the scale of millions or even billions. m ) combined with the SDL model, the size of the feature vector can be easily increased dramatically. Its purpose is to surpass the size of deep learning datasets, thereby producing application effects that go beyond deep learning.

[0115] In practical applications, facial image data often only contains one photo obtained from an ID card. Therefore, how to use this photo to generate a probability distribution database with the maximum probability value and the maximum probability scale is the key to improving facial recognition by importing the SDL model.

[0116] The present invention proposes to use various image processing methods that can improve the image recognition effect, transform a facial image into a plurality of expanded images of facial images produced by different environments that can simulate the sample image, and use the plurality of expanded images as a set of sample data for the image; and a method for making an image also generate a probability distribution of sample feature values.

[0117] The specific image processing method proposed here is only an example. It is within the scope of the present invention to transform an image into multiple images by using various image processing methods that can improve image recognition effects and to generate a probability distribution of a sample feature vector for image recognition.

[0118] The following is an image processing method for grayscale adjustment.

[0119] like Figure 7 (a) is the original grayscale value of any 3*3 pixel in the original face image. Figure 7 (b) Exchange the maximum grayscale value and the center grayscale value in the original grayscale value of 3*3 pixels. Figure 7 (c) Exchange the minimum grayscale value and the center grayscale value in the original grayscale values ​​of 3*3 pixels. Figure 7 (d) Each gray value in the original gray value of 3*3 pixels is self-organized using the probability scale to replace the central gray value with the high probability value. Figure 7 (e) Exchange the maximum grayscale value and the center grayscale value in the original grayscale values ​​of 3*3 pixels.

[0120] Following the above method, you can also Figure 7 (a) The diagonal line and the center cross line of the original grayscale value of 3*3 pixels are replaced by the maximum grayscale value and the minimum grayscale value in the original grayscale value of 3*3 pixels, and the maximum probability value is obtained by self-organization using the probability scale.

[0121] Figure 8 It is an image processing method that emphasizes border information.

[0122] like Figure 8 As shown in (a); the image is derived in the x and y directions respectively, and then Figure 8 The constants in the left and right 3*3 grids of (a) are multiplied according to the corresponding pixels, and the grayscale value of the original pixel is replaced by the result. Figure 8 As shown in (b); the image is derived in the x and y directions respectively, and then Figure 8 The grayscale value of the original pixel is replaced by the result of multiplying the constants in the left and right 3*3 grids of (b).

[0123] Figure 9It is another image processing method that emphasizes border information.

[0124] same Figure 8 Same, such as Figure 9 As shown in (a), by multiplying the x-direction derivative result with this template, the processing effect of the horizontal border filter can be obtained. Figure 9 As shown in (b), y The directional derivative result is multiplied by this template to obtain the processing effect of the horizontal border filter.

[0125] In addition to transforming a face image into multiple images, it is also important to emphasize that the quality of the image can be improved.

[0126] In image processing, in order to improve the visual quality of the image, image brightness transformation, color transformation, image enhancement, suppression of certain components, geometric transformation, etc. are used.

[0127] In addition, by strengthening the high-frequency components of the image, the facial features in the image can be made clear and the details obvious, and by strengthening the low-frequency components of the image, the noise in the image can be removed.

[0128] The image can also be divided into multiple images through image filtering, which can improve the scale of the feature vector while improving the image quality.

[0129] For example, the grayscale value transformation functions include: logarithmic transformation, gamma transformation, piecewise linear transformation, histogram transformation, etc.

[0130] Spatial filtering includes: smoothing spatial filtering, statistical sorting filtering, sharpening spatial filtering, unsharpening concealment and high-lift filtering, gradient sharpening, etc.

[0131] Frequency-based filtering includes: ideal low-pass filtering, ideal high-pass filtering, Butel low-pass filtering, Butel high-pass filtering, Gaussian low-pass filtering, Gaussian high-pass filtering, passivation mask, high-boost filtering, high-frequency emphasis filtering, homomorphic filtering, selective filtering, and 2-trap filtering.

[0132] The processing of noise probability density function includes: Gaussian noise, Rayleigh noise, Irish noise, exponential noise, uniform noise, bibasic impulse noise, etc.

[0133] Spatial filtering to restore noise includes: mean filter, arithmetic mean filter, geometric mean filter, harmonic mean filter, inverse harmonic mean filter, statistical sorting filter, adaptive filtering, frequency domain elimination of periodic noise, etc.

[0134] In order to improve the accuracy of face recognition, the present invention also proposes the following method related to recognition conditions.

[0135] Figure 10It is a schematic diagram of a single photo image versus a single recognition image.

[0136] like Figure 10 As shown in the figure, a facial image of more than 10,000 people in the database is used, one for each person, and a feature vector (PV) is extracted using the above-mentioned image transformation method or the facial features position method, or using two or more methods at the same time. max1～n ), log the feature vector, then use the camera to take a face image of the logged-in photo, and use the above-mentioned image transformation method or the method of facial features position, or use two methods at the same time, or more methods to extract a feature vector (sp_PV max1～n ) for face recognition.

[0137] Here, a method for performing positive recognition on images is proposed.

[0138] Figure 11 This is a schematic diagram of the recognition of a single-shot image using simulated multi-scene photo images.

[0139] like Figure 11 As shown in the figure, the above image processing method is used to simulate images taken in multiple scenes for each person in the database of more than 10,000 face images. Then, the above image transformation method or the method of facial features position are used, or two or more methods are used at the same time to extract a set of feature vectors, and the probability scale self-organization processing of the feature vectors is performed to obtain the maximum probability value (PV) generated by the probability scale self-organization. max1～n ) is the expected value of the maximum probability space in statistics, the scale of the maximum probability space (PS max1～n ) is the variance of the maximum probability space, and is logged into the database. Then, a face image of the logged-in photo is captured with a camera, and a feature vector (sp_PV) is extracted using the above-mentioned image transformation method or the method of facial features position, or using both methods, or more methods. max1～n ) for face recognition.

[0140] Alternatively, by directly taking multiple photos of a person, we can obtain images of several different scenes, and then use the above-mentioned image transformation method or the method of facial features position, or use two or more methods at the same time to extract a set of feature vectors, perform probability scale self-organization processing on the feature vectors, and obtain the maximum probability value (PV) generated by the probability scale self-organization. max1～n ), the scale of the maximum probability space (PS max1～n ) is the variance of the maximum probability space, and is logged into the database. Then, a face image of the logged-in photo is captured with a camera, and a feature vector (sp_PV) is extracted using the above-mentioned image transformation method or the method of facial features position, or using both methods, or more methods. max1～n) for face recognition.

[0141] Here, a method for reverse recognition of images is proposed.

[0142] Figure 12 It is a schematic diagram of image recognition of a single photo in multiple scene shooting.

[0143] like Figure 12 As shown in the figure, for each face image in the database of more than 10,000 people, a feature vector (PV) is extracted using the above-mentioned image transformation method or the method of facial features position, or using two or more methods at the same time. Max1～n ) to log the feature vector. Then, identify the face image of a real person from more than 10,000 face images in the database. Use a camera to continuously shoot the face of the person to be identified to obtain several face images of the person to be identified. Then, use the above-mentioned image transformation method or the method of facial feature position, or use two or more methods at the same time to extract a set of feature vectors. Perform the probability scale self-organization processing of the feature vectors to obtain the maximum probability value (sp_PV) generated by the probability scale self-organization. max1～n ), that is, the expected value of the maximum probability space in statistics, the scale of the maximum probability space (sp_PS max1～n ) is the variance of the maximum probability space, and face recognition is performed. This method can also be called the reverse recognition method.

[0144] Here, a method for performing bidirectional recognition on images is proposed.

[0145] Figure 13 Schematic diagram of single photo image recognition for multi-scene shooting and simulated multi-scene shooting

[0146] like Figure 13 As shown in the figure, a camera is used to shoot several faces of the identified persons at the same time or in different scenes, and several face images of the identified persons are formed. Alternatively, more than 10,000 face images in the database are used, and one image is taken for each person. The above-mentioned image processing method is used to simulate images shot in multiple scenes, and several images simulating different scenes can be obtained. Then, the above-mentioned image transformation method or the method of the position of the facial features are used, or two or more methods are used at the same time to extract a set of feature vectors, and the probability scale self-organization processing of the feature vectors is performed to obtain the maximum probability value (PV) generated by the probability scale self-organization. max1～n ) That is, the scale of the maximum probability space (PS max1～n ) is the maximum probability variance logged into the database.

[0147] Then, a real person is identified from more than 10,000 facial images in the database. The face of the person to be identified is continuously photographed with a camera to obtain several facial images of the person to be identified. Then, a set of feature vectors is extracted using the above-mentioned image transformation method or the method of facial features position, or two or more methods at the same time. The probability scale self-organization processing of the feature vector is performed to obtain the maximum probability value (sp_PV) generated by the probability scale self-organization. max1～n ), the scale of the maximum probability space (sp_PS max1～n ) is the maximum probability variance, and face recognition is performed. This method can also be called a two-way recognition method.

[0148] The following proposes a method for segmenting the probability distribution of the eigenvalues ​​of the login feature vector for quantitative pattern recognition.

[0149] Figure 14 It is a schematic diagram of dividing a probability space into multiple probability spaces.

[0150] like Figure 14 As shown; Figure 14 (a) is a complete probability space Φ of eigenvalues, which can be divided into the following categories according to the scale of the maximum probability space: σ'3~σ'2, σ'2~-σ'1, σ'1~σ1, σ1~σ2, σ2~σ3, Figure 14 (b) The five probability spaces Φ1, Φ2, Φ3, Φ4 and Φ5, each of which can generate five maximum probability values ​​μ1, μ2, μ3, μ4 and μ5, as well as the scale of the maximum probability space, through the processing of probability scale self-organizing machine learning.

[0151] Here, the probability distribution value of the Φ3 probability space is 68.2%, the probability distribution value of the Φ2 and Φ4 probability spaces is 27.2%, and the probability distribution value of the Φ1 and Φ5 probability spaces is 4.3%.

[0152] exist Figure 14 In a complete probability space of (a), if the probability distribution value is only 4.3% of the recognition results appear, it will also be identified as belonging to Figure 14 (a) The probability space corresponding to the recognition result is prone to errors in the recognition result. The probability distribution of a feature value is divided into 5 probability distributions. During recognition, if the sample feature value falls in the probability space of Φ1 and Φ5, the recognition result can be rejected. By re-identifying, a recognition result with a higher probability value can be obtained.

[0153] For the probability distribution of the recognition result of a feature vector, each eigenvalue belonging to the sample feature vector can be assigned to the probability distribution value of the reliability or quality evaluation result of the corresponding feature vector. Then, this value can be processed by machine learning using a self-organizing probability scale to obtain the maximum probability value of the recognition result of the feature vector. According to the probability distribution value of the maximum probability space where the sample data of the recognition result falls, the probability distribution value of the recognition result can be obtained. Based on the large or small probability distribution value, it can be judged whether to reject the recognition.

[0154] Next, a method for calculating the accuracy of the recognition result is proposed for the learned and registered feature vector data.

[0155] Assume that there are m eigenvectors in the database, each with n eigenvalues. After t training cycles, the data obtained can be used to self-organize machine learning using probability scales to obtain the following maximum probability matrix P: (Formula 6)

[0156]

[0157] And the scale matrix M of the maximum probability space:

[0158] (Formula 7)

[0159]

[0160] The above two matrices can be analyzed using the (EPD distance) formula that can unify the distance scale of Euclidean space and probability space.

[0161] Figure 15 is a diagram of two probability spaces in Euclidean space.

[0162] like Figure 15 As shown: (∈) is a Euclidean space covering two probability spaces. There are two probability spaces in Euclidean space as well as (ρ 1 ) is the probability space The maximum probability value. (ρ 2 ) is the probability space The maximum probability value of . In addition, (σ 1 ) is the probability space The scale of the maximum probability space, (σ 2 ) is the probability space The scale of the maximum probability space, the probability space The probability value of the first maximum probability space Probability Space The probability value of the second maximum probability space Probability Space The probability value of the third maximum probability space Probability Space The probability value of the first maximum probability space Probability Space The probability value of the second maximum probability space Probability Space The probability value of the third maximum probability space

[0163] Furthermore, let the probability space as well as The maximum probability value is (ρ 1 ) and (ρ 2 ) are the elements w of the two data sets j ∈W and v j ∈V. Let the maximum probability value be (ρ 1 ) and (ρ 2 ) is connected by a straight line, and there is a sample data β at any point in the middle of the straight line j ∈B, find the sample data set β j ∈B is the data set W or V belongs to, which can be obtained by the following EPD distance formula:

[0164] Formula 8

[0165]

[0166]

[0167] The distance between the unified Euclidean space and the probability space satisfies the following conditions:

[0168] (1) Non-negativity: v, d(w, v) 20;

[0169] (2) Non-degeneracy: d(w, v) = 0, then w = v;

[0170] (3) Symmetry: v, d(w, v) = d(v, w);

[0171] (4) Triangle inequality: r,vd(w,v)≤d(w,r)+d(r,v).

[0172] In order to improve the accuracy of face recognition, a more accurate scale can be introduced that can fully take into account fuzzy information and probability information. A method for defining the probability measure of fuzzy events across different spaces is characterized by: while considering the spatial distance between data, the probability value of the probability distribution of the data in the probability space must also be considered.

[0173] Assume that the set W belongs to the probability space. The formula of fuzzy event probability measure (FPM measure) of set V can be derived from the following formula:

[0174] Formula 9

[0175]

[0176] Here, pf j (vj ) set elements w j ∈W(j=1,2,...,n) is in the set element v j The probability distribution value of the position in the probability space of ∈V(j=1, 2, ..., n);

[0177] Similarly, pf j (wj) is the set element v j ∈V(j=1,2,...,n) is in the set element w j The probability distribution value of the position in the probability space ∈W(j=1, 2, ..., n).

[0178] d max It is w j to v j The maximum distance.

[0179] The membership function formula for the fuzzy event probability measure can include any fuzzy numerical value between 0 and 1 that can be obtained by constructing the objective function according to certain artificial rules. There are various methods for defining similar official script functions, and all methods involving the definition of fuzzy event probability measures fall within the scope of the present invention.

[0180] The spatial information of the above fuzzy event probability measure is based on the distance between the unified Euclidean space and the probability space, and satisfies the following distance conditions:

[0181] (1) Non-negativity: v, F(w, v) ≥ 0:

[0182] (2) Non-degeneracy; F(w, v) = 0, then w = v;

[0183] (3) Symmetry: v, F(w, v) = F(v, w);

[0184] (4) Triangle inequality: r, v F (w, v) ≤ F (w, r) ∪ F (r, v).

[0185] Here, in order to improve the accuracy of face recognition, a method is proposed that considers the ratio between the minimum distance between eigenvectors and the cumulative value of the scale of the maximum probability space of each eigenvalue of the eigenvector.

[0186] Using the above EPD distance formula, we can get the distance matrix between feature vectors:

[0187] (Formula 11)

[0188]

[0189] Here, let d ij =d ji , d ii =d jj =0.

[0190] Using the above FPM measure formula, we can get the measure matrix between eigenvectors:

[0191] (Formula 12)

[0192]

[0193] Here, let d ij =d ji , d ii =d jj = 0, and f ij =f,f ii =f jj =0.

[0194] Here, two eigenvectors corresponding to the minimum EPD distance can be obtained, or two eigenvectors corresponding to the maximum FPM measure can be obtained.

[0195] Then calculate the two closest EPD distance values ​​D min , and the maximum probability space scale of the two eigenvectors corresponding to the two closest EPD distance values and The cumulative value of:

[0196] (Formula 13)

[0197]

[0198] According to Formula 10 and Formula 12, it can be concluded that the necessary and sufficient conditions for face recognition to achieve the highest recognition accuracy are:

[0199] (Formula 14)

[0200] D min ≥M min or

[0201] Here we can quantitatively calculate the accuracy of face recognition through the data set of the logged feature vector. With such a result, if we need to improve the accuracy of face recognition, we must first solve the quality of feature information extraction of face images. The higher the quality, the better. min With M min The larger the ratio is, the smaller it is or even less than 1, the worse the recognition accuracy is.

[0202] Here, a method is proposed that considers the effect of a large-scale registered feature set on the recognition result and can improve the quality of the feature vector.

[0203] To improve the quality of the eigenvector, that is, to increase the value of (Formula 14), the image transformation algorithm used must be selected, and some image transformation algorithms that can achieve higher eigenvector quality must be adopted. After determining an image transformation algorithm, the average distance method is used to perform probabilistic scale self-organizing machine learning on the set of scales of the maximum probability space of each eigenvalue contained in the eigenvector. Using the scale of the maximum probability space of reliable, high-quality eigenvalues ​​obtained from the probabilistic scale self-organizing machine learning process as a benchmark, eigenvalues ​​are filtered according to the scale, retaining those that meet the scale and eliminating those that do not. This results in a set of high-reliability, high-quality eigenvalues. This set is used as the final eigenvector for face recognition. A binary code is used to indicate which eigenvectors are high-reliability, high-quality eigenvalues ​​and which are not, indicating which are high-reliability, high-quality eigenvalues ​​and which are not. This code is then stored in the database. Furthermore, since each eigenvector has a different number of high-reliability, high-quality eigenvalues, the average distance method is used to unify the scales of the eigenvectors.

[0204] In order to improve the accuracy of face recognition, a method is proposed here that considers the common probability distribution information of complex eigenvalues, as well as a method that considers the common probability distribution information of sample eigenvalues ​​at the common position of the common probability distribution of the logged complex eigenvalues ​​and the probability distribution of the logged eigenvalues.

[0205] Figure 16 It is a schematic diagram of the pattern recognition method that uses common probability distribution information.

[0206] like Figure 16As shown in the figure, in pattern recognition, it is often necessary to calculate which of the two maximum probability eigenvalues ​​ω4 and ω8 is closest to the sample eigenvalue τ. Using the above EPD distance formula and the FPM measure formula, this can be directly calculated. However, when calculating the distance between ω4 and τ or the FPM measure, the common probability distribution of ω4 and ω8 is not considered. When training data here, a matrix of the common maximum probability values ​​between the elements of the eigenvector can be obtained:

[0207] (Formula 15)

[0208]

[0209] The matrix of the scale of the common maximum probability space between the elements of the eigenvector can also be obtained:

[0210] (Formula 16)

[0211]

[0212] While training the maximum probability value of the feature vector and the scale of the maximum probability space, we also train the common maximum probability value of the two feature vectors and the scale of the common maximum probability space, such as Figure 16 As shown; let φ1 and φ3 be the probability distribution of any independent eigenvector in any two different eigenvectors, φ2 be the common probability distribution of φ1 and φ3, ω4 be the position corresponding to the maximum probability value of the probability distribution φ1, ω8 be the position corresponding to the maximum probability value of the probability distribution φ3, ω6 be the position corresponding to the maximum probability value of the common probability distribution φ2 of φ1 and φ3, from (Formula 8) we can get the sample eigenvalue τ corresponding to φ1 and φ3, the result is that the EPD distance from the position of τ to the position of ω4 is ε1, and the EPD distance from the position of τ to the position of ω6 is ε2, then the new distance formula from the position of τ to the position of ω4 taking into account the common probability distribution information of the eigenvalue is:

[0213] (Formula 17)

[0214]

[0215] The new distance formula from the position of τ to the position of ω8, which also takes into account the common probability distribution information of the eigenvalues, is:

[0216] (Formula 18)

[0217]

[0218] Here, we can also assume that the sample image is also a plurality of images read from different environments. Each eigenvalue of the sample feature vector can also generate a probability distribution. In this case, the probability distribution of the eigenvalue of the sample feature vector can be regarded as φ3. To determine whether φ1 and φ3 belong to the same probability distribution, it is easier to obtain the best recognition result by using the common probability distribution information φ2 of φ1 and φ3. In this case, ε1 = ω6-ω4, ε2 = ω8-ω6, and ε3 = ω8-ω4, then

[0219] (Formula 19)

[0220]

[0221] The larger the value of , the closer the relationship between φ1 and φ3.

[0222] Suppose the probability distribution value of the probability distribution φ1 of the eigenvector where the sample eigenvalue τ is located is δ1, the probability distribution value of the common probability distribution φ2 of the two eigenvectors where the sample eigenvalue τ is located is δ2, similarly, the probability distribution value of the probability distribution φ3 of the eigenvector where the sample eigenvalue τ is located is δ3, then after introducing the common probability distribution information of the two eigenvalues, the probability distribution value of the sample eigenvalue τ at the position of the probability distribution φ1 of the eigenvector should be:

[0223] (Formula 20)

[0224] P(φ1)=P(δ1∩δ2)=P(δ1)P(δ2 / δ1)

[0225] After introducing the common probability distribution information of the two eigenvalues, the probability distribution value of the sample eigenvalue τ at the position φ3 of the probability distribution of the eigenvector should be:

[0226] (Formula 21)

[0227] P(φ3)=P(δ3∩δ2)=P(δ3)P(δ2 / δ3)

[0228] After introducing the common probability distribution information of the two eigenvalues, the new fuzzy event probability measurement formula between τ and ω4 is:

[0229]

[0230] After introducing the common probability distribution information of the two eigenvalues, the new fuzzy event probability measurement formula between τ and ω8 is:

[0231] (Formula 23)

[0232]

[0233] In order to improve the accuracy of face recognition, the probability information of fuzzy event probability measurement is considered in addition to the sample eigenvalue s j Falling to the eigenvalue f corresponding to a certain login eigenvector j The probability distribution area of ​​the region, the probability distribution value of the region. Another is to introduce the common probability distribution information of the two eigenvalues. More importantly, consider the eigenvalues ​​s of the sample eigenvector j (j=1, 2, ..., n), each eigenvalue f of the corresponding eigenvector falls into j The probability distribution value on the probability distribution area of ​​(j=1, 2, ..., n), or the probability value obtained after introducing the common probability distribution information of each eigenvalue of the two eigenvectors.

[0234] Here, we propose to consider the eigenvalue s of the sample eigenvector S j ∈S(j=1,2,…,n) belongs to a registered feature vector F i (w) The corresponding eigenvalue f ij (w) A method for defining probability information with the maximum probability.

[0235] Figure 17 It is a schematic diagram of a method for defining a new probability value taking into account the overall probability.

[0236] Let the eigenvalue s of the sample eigenvector S be j ∈S(j=1,2,…,n), respectively, in a certain registered feature vector W i The corresponding eigenvalue w ij The probability value of the position on the probability distribution is pf ij (j=1,2,…,n), self-organize the pf by probability scale ij (j=1, 2, ..., n) performs machine learning to obtain a maximum probability value pf imax (w) , take this probability value as the sample characteristic value s j and eigenvalue f ij (w) The probability value of the membership degree is considered from the overall perspective of the relationship between the two feature vectors, so the recognition accuracy of the image can be improved.

[0237] From this we can get the sample feature vector S, which belongs to a certain registered feature vector W i The fuzzy event probability measure is:

[0238] (Formula 24)

[0239]

[0240] Here, σ1 It is w ij The maximum probability scale of the probability space, σ 2 It is v ij The scale of the maximum probability of the probability space. From (Formula 23), it can be seen that if the feature vector S→W i Then FPM i (w) →100, whereas FPM i (w) →0.

[0241] Similarly, let the eigenvalue s of the sample eigenvector S be j ∈S(j=1,2,…,n), respectively, in a certain registered feature vector V i The corresponding eigenvalues ​​v ij The probability value of the position on the probability distribution is pf i j (v) (j=1,2,…,n), self-organize the pf by probability scale ij (v) (j=1, 2, ..., n) performs machine learning to obtain a maximum probability value pf i_max (v) You can also use pf i_max (v) Value as sample eigenvalue s j ∈S(j=1,2,…,n) belongs to all eigenvalues ​​v ij The probability value of .

[0242] According to formula 23, the sample feature vector S belongs to a certain registered feature vector V i The fuzzy event probability measure is:

[0243] (Formula 25)

[0244]

[0245] From (Formula 24), it can be seen that if S→V i Then FPM i (v) →100, whereas FPM i (v) →0.

[0246] With the help of (Formula 17), (Formula 18) and (Formula 19), s j -v ij , s j -w ij and w ij -v ij, can be transformed into a distance that takes into account the common probability distribution information of the eigenvalues. By substituting the probability information definitions of (Formula 20) and (Formula 21) into the above distance formula, we can obtain the solution to the fuzzy event probability measure. We can also directly use (Formula 22) and (Formula 23) to directly construct the solution to the fuzzy event probability measure.

[0247] The size of the feature vector set should be large. The number of images for face recognition increases. For example, the face is divided into several regions according to the location of the facial features. Each additional region will increase the size of the feature vector exponentially. The increase and decrease of the feature vector data set using the SDL model has a complexity of 0(n 2 ) is close to linear, and the size of the feature vector dataset can be increased infinitely.

[0248] There are two ways to increase the feature vector dataset infinitely; one is Figure 6 As shown in the figure, an image transformation method is used to map the image to a larger feature dataset. On the other hand, the image is infinitely segmented, and each segment corresponds to an image transformation, thereby obtaining a huge feature dataset, which can achieve the highest recognition results. This method is not only suitable for face recognition and image recognition, but also for voice recognition and other applications.

[0249] Here we give a machine learning model of probability scale self-organization, and the given probability distribution sequence p1, p2, ...p m p i ∈G, the element (i=1, 2, ..., m) of the set, its maximum probability value is A(G), and the scale of the maximum probability space of the maximum probability value A(G) is M[G, A(G)]. Then, according to the n-1th self-organizing iterative process, the maximum probability value A(G) in the calculated (n-1) )'s maximum probability space scale M[G (n-1) ,A(G (n-1) )] If there are k probability distribution series p (n) 1, p (n) 2, ...p (n) k , p (n) i ∈G (n) (i=1, 2, ..., k), then the result of the self-organization of the nth probability scale is:

[0250] (Formula 26)

[0251] A (n) =A(G (n) )

[0252] M (n) =M[G(n) , A(G (n) )]

[0253] G (n) =G{A(G (n-1) ), M[G (n-1) , A(G (n-1) )]}

[0254] The iterative results of the above probability scale self-organized machine learning model can be used for a given sequence of random information p1, p2, ...p m After iterative processing, we can obtain a maximum probability value that can represent this set of number series, the maximum probability space that exists in this number series, and the scale that can represent the maximum probability of this space.

[0255] When conducting self-organizing machine learning of probability scales in two-dimensional space or three-dimensional space, it can be achieved by simply replacing the maximum probability value of the two-dimensional space or three-dimensional space with the mechanical center of gravity of the two-dimensional space or three-dimensional space and the scale of the maximum probability space of the two-dimensional space or three-dimensional space with the two-dimensional or three-dimensional variance according to the above-mentioned iterative method.

[0256] When performing multi-objective processing on an objective function, multiple objectives often aim to achieve the optimal solution in the maximum probability space. In particular, multiple objectives have a multiplicative property that mutually promotes the acquisition of the optimal solution. Furthermore, when field feature information is required, since they are in a common maximum probability space, they automatically promote the generation of field features, achieving automatic machine learning without training. Furthermore, the iterative process in the maximum probability space also promotes the convergence of multiple objectives towards the optimal solution. This paper proposes multi-objective automatic machine learning.

[0257] Here is an application example of image approximation. The above-mentioned probability scale self-organizing machine learning is introduced. In the image of the maximum probability space for image approximation, image processing required for image approximation, such as binarization, can be performed. Because the range of the processed image can be narrowed to an image closer to the approximated image, a more ideal image processing result will be obtained. This is more conducive to the next iteration to obtain an image that is closer to the maximum probability space, and a solution that is closer to the image can be obtained. There is no need to perform prior machine learning on the image to be approximated, and the optimal solution for image approximation can be automatically obtained.

[0258] Figure 18 It is a process flow chart for multi-purpose automatic machine learning.

[0259] like Figure 18As shown in the flowchart of multi-purpose automatic machine learning, it needs to go through the initialization step S1, the step of obtaining the maximum probability space S2, the processing step of purpose 1 S3, and the step of 3+1 Perform the processing steps of purpose 2, ..., S 3+n The process comprises the steps of processing the target n, S4 judging step and S5 ending step.

[0260] Initialization step S1: Set the initial conditions for the iteration, such as the initial image processing method and the initial input probability distribution data. This can be used to obtain the initial maximum probability spatial scale of the data, or directly provide an estimated maximum probability spatial scale. The evaluation function for the iteration effect, to prevent infinite iterations, sets the maximum number of iterations, etc.

[0261] Obtaining the maximum probability space step S2: For a maximum probability space, the scale of the maximum probability space of the space can be obtained. This scale can be used to screen out a space with a greater probability. After obtaining a new maximum probability space, a new scale of the maximum probability space can be generated in the new maximum probability space, which can be used to generate a maximum probability space with a greater probability in the next iteration.

[0262] Perform processing step S3 for purpose 1: According to the dependency relationship between the processing results of the multi-purpose objective functions in the maximum probability space, select the objective function of purpose 1 with the highest priority to be processed in this step.

[0263] Perform processing step S of purpose 2 3+1 : After performing the processing steps of purpose 1, conditions that are more suitable for obtaining the optimal solution are provided for performing the processing steps of purpose 2, so that the objective function of purpose 2 obtains a more prominent optimal solution.

[0264] …

[0265] Perform processing step S for purpose n 3+n : After performing the processing steps of purpose n-1, the most suitable conditions for obtaining the optimal solution are provided for performing the processing steps of purpose n, so that the objective function of purpose n finally obtains the most outstanding optimal solution.

[0266] Judgment step S4: judge whether the iteration result is completed through the evaluation function of the iteration effect and the maximum number of iterations. If it is not completed, jump to the maximum probability space step S2. If it is completed, go to the next step.

[0267] End step: The program ends, or returns to the main program.

[0268] Radar sensors play a crucial role in autonomous driving. However, current radar applications simply read radar data to determine obstacles around the autonomous vehicle. Because radar information is discrete and each piece of data has significant random characteristics, machines cannot directly understand it. Therefore, the reliability of obstacle detection based on radar data is very low.

[0269] Here, we propose a method for processing radar data that can be understood by machines.

[0270] Figure 19 It is a flow chart of the method for machine understanding radar data processing.

[0271] like Figure 19 As shown, the method for machine understanding radar data processing is divided into: obtaining radar data step S1, performing function interpolation step S2, performing video transformation step S3, and image recognition step S4.

[0272] Obtain radar data step S1: Obtain data after the radar completes a scan of surrounding obstacles, generally 16-line data, that is, data composed of 16 lines that can reflect the distance of surrounding obstacles.

[0273] Performing function interpolation step S2: Increasing the number of radar lines can improve the accuracy of observing surrounding obstacles, but the cost increases exponentially with the number of lines increased. How to use lower radar accuracy to achieve higher precision application effects? Here, it is proposed that in order to improve the accuracy of radar data and achieve higher line effect with fewer radar lines, radar data can be interpolated by function. Commonly, B-spline function, ordinary spline function, Lagrange interpolation, Hermite interpolation, trigonometric polynomial interpolation, Gaussian trigonometric interpolation, Sink interpolation, multiple regression, least squares method, neighboring point interpolation, bilinear interpolation, bisquare interpolation, bicubic interpolation, cubic interpolation, etc. can be used.

[0274] The video transformation step S3 is performed. In this step, the discrete radar data is converted into higher-precision radar data through function interpolation. After the video transformation, a radar video image with the distance to the obstacle as the grayscale value can be obtained.

[0275] Image recognition step S4: The radar video image obtained in the previous step is similar to the distance video image obtained by the image sensor, which uses image distance as the grayscale value. Various obstacles are located at different distances, making it easy to separate the various obstacles. Furthermore, the various obstacles are distributed at corresponding positions in the image space with a high density, or what can be considered a high-probability distribution. This distribution can be a three-dimensional conical distribution or a three-dimensional rectangular distribution. By utilizing this characteristic and introducing the above-mentioned two-dimensional or three-dimensional space probability scale self-organization, obstacles with a three-dimensional conical distribution or a three-dimensional rectangular distribution can be directly identified for machine understanding.

[0276] Two identical images captured at a certain interval by a binocular camera will be misaligned. The magnitude of this misalignment can be used to determine the distance of an object. Alternatively, a monocular camera can utilize the changing distance as a car moves, taking advantage of the misalignment information in the images captured front and back. This information can be converted into distance information. Therefore, a binocular or monocular camera can also produce a grayscale distance image. This also makes it easy to separate various obstacles, and obstacles can be distributed in a 3D conical or rectangular shape in distance space. Obstacles with 3D conical or rectangular distributions can also be directly identified for machine understanding.

Claims

1. A method for calculating face recognition accuracy, characterized by: The camera captures several faces to be recognized simultaneously or in different scenes, extracts a set of feature vectors, and performs a probability scale self-organization process on the feature vectors to obtain the maximum probability value generated by the probability scale self-organization and the scale of the maximum probability space, i.e., the maximum probability variance; In order to improve the accuracy of face recognition, fuzzy information and probability information are fully taken into account. A definition method of fuzzy event probability measurement across different spaces is introduced. While considering the spatial distance between data, the probability value of the probability distribution of the data in the probability space must also be considered. Assume that the set W belongs to the probability space. The formula of the fuzzy event probability measure of the set V, namely the FPM measure, can be derived from the following formula: Here, pf j (vj) is the set element w j ∈W(j=1,2,...,n) is in the set element v j The probability distribution value of the position in the probability space of ∈V(j=1, 2, ..., n); Similarly, pf j (wj) is the set element v j ∈V(j=1,2,...,n) is in the set element w j The probability distribution value of the position in the probability space of ∈W(j=1, 2, ..., n); d max It is w j to v j The maximum distance; The spatial information of the above fuzzy event probability measure is based on the distance between the unified Euclidean space and the probability space, and satisfies the following distance conditions: (1) Non-negativity: (2) Non-degeneracy: F(w, v) = 0, then w = v; (3) Symmetry: (4) Triangle inequality: In order to improve the accuracy of face recognition, a method is proposed that considers the ratio between the minimum distance between eigenvectors and the cumulative value of the scale of the maximum probability space of each eigenvalue of the eigenvector; Using the EPD distance formula, we can get the distance matrix between eigenvectors: Here, let d ij =d ji , d ii =d jj =0; Using the above FPM measure formula, we can get the measure matrix between eigenvectors: Here, let d ij =d ji , d ii =d jj = 0, and f ij =f ji , f ii =f jj =0; Here we can get the two eigenvectors corresponding to the minimum EPD distance, or we can get the two eigenvectors corresponding to the maximum FPM measure; Then calculate the two closest EPD distance values ​​D min , and the maximum probability space scale of the two eigenvectors corresponding to the two closest EPD distance values and The cumulative value of: The necessary and sufficient conditions for face recognition to achieve the highest recognition accuracy are: D min ≥M min or Here, the accuracy of face recognition can be quantitatively calculated by the data set of the logged feature vector. If the accuracy of face recognition needs to be improved, the quality of feature information extraction of face images must be improved first. The higher the quality, the better. min With M min The larger the ratio is, the smaller it is or even less than 1, the worse the recognition accuracy is. Among them, let the probability space as well as The maximum probability value is (ρ 1 ) and (ρ 2 ), probability space as well as The elements w of the two data sets j ∈W and v j ∈V, and then assume that the maximum probability value is (ρ 1 ) and (ρ 2 ) is connected by a straight line, and there is a sample data β at any point in the middle of the straight line j ∈B, find the sample data set β j Belonging to the data set W or V is obtained by the following EPD distance formula: The distance between the unified Euclidean space and the probability space satisfies the following conditions: (1) Non-negativity: (2) Non-degeneracy: d(w, v) = 0, then w = v; (3) Symmetry: (4) Triangle inequality: (∈) is a Euclidean space covering two probability spaces. In the Euclidean space, there are two probability spaces as well as (ρ 1 ) is the probability space The maximum probability value of (ρ 2 ) is the probability space The maximum probability value of 1 ) is the probability space The scale of the maximum probability space, (σ 2 ) is the probability space The scale of the maximum probability space, the probability space The probability value of the first maximum probability space Probability Space The probability value of the second maximum probability space Probability Space The probability value of the third maximum probability space Probability Space The probability value of the first maximum probability space Probability Space The probability value of the second maximum probability space Probability Space The probability value of the third maximum probability space

Images