• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A pattern recognition system and method that generates a feature vector by multiplying an image vector with a hollow matrix. The hollow matrix is generated from a Gabor function which is a sinusoidal wave multiplied by a Gaussian function. The Gabor function is a function of a set of parameters comprising a parameter relating to the direction of the sine wave, a parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave. The wavelength takes at least two values, with a first wavelength value less than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first wavelength value is less than one. second wavelength value and greater than or substantially equal to half of the second wavelength value.
    公开号:BE1025502B1
    申请号:E2015/5241
    申请日:2015-04-15
    公开日:2019-03-27
    发明作者:Frédéric Collet；Jordi Hautot；Michel Dauw；Meulenaere Pierre De；Olivier Dupont；Günter Hensges
    申请人:I.R.I.S.；
    IPC主号:
    专利说明:

    PATTERN RECOGNITION SYSTEM AND METHOD USING GABOR FUNCTIONS
    FIELD OF THE INVENTION
    The present invention relates to a pattern recognition system. More specifically, the present invention relates to a pattern recognition system using a Gabor function.
    BACKGROUND OF THE INVENTION
    A pattern recognition system can be an optical character recognition (OCR) system. OCR systems are known. They convert the image of the text into machine readable code using a character recognition process. In an OCR system, images of what could be characters are isolated and a character recognition process is used to identify the character.
    Known optical character recognition processes generally include:
    · A normalization step which generates a normalized matrix from an input image;
    · A feature extraction step; and · a classification step to identify the character.
    The characteristic extraction step generates a characteristic vector which characterizes the input image and the classification step identifies the character on the basis of this characteristic vector. In some OCR processes, the feature extraction step includes filtering with a Gabor filter. The choice of the Gabor filter is decisive for the OCR process because the Gabor filter
    BE2015 / 5241 determines the characteristic vector to identify the character. The feature vector should contain the information necessary to identify the trait with great accuracy. Too large a feature vector makes calculations slow and a too small feature vector decreases the accuracy of character identification. The known OCR methods using Gabor filters are too slow or have too low an accuracy. This is particularly important for the identification of Asian characters because of the extremely high number of Asian characters. Another disadvantage of the known Gabor filters lies in the fact that they do not function adequately with the subsequent classification step.
    US 7174044 B2 discloses a known method of character recognition based on Gabor filters which extract information from the specific directions of the characters. This method uses an average over Gabor filter regions and includes many calculations and a large vector of characteristics. This makes OCR processes using this process too slow.
    The document “High performance Chinese OCR based on Gabor features, discriminative feature extraction and model training” by Qiang Huo, Yong Ge and Zhi-Dan Feng in the Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, 2001, Volume 3, describes a known OCR process for Chinese characters. This process is fast but the accuracy of the recognition is not extremely high. .
    The dissertation titled “Chinese OCR System Based on Gabor Features and SVM” by Dai Wei at Shanghai Jiaotong University describes another OCR process. SVM stands for Support Vector Machine, a supervised learning model that uses learning algorithms
    BE2015 / 5241 associated for data analysis and recognition algorithms. Such an SVM system requires a very large set of learning which makes it impractical or inaccurate.
    SUMMARY OF THE INVENTION
    It is an object of this invention to provide a pattern recognition method which allows rapid and accurate identification of the pattern.
    Another object of this invention is to provide a computer program product for implementing said pattern recognition system.
    These objects are achieved according to the embodiments of the invention.
    In one embodiment, the method of identifying a shape in an input image comprises the steps consisting in:
    a) normalizing the input image into a standardized matrix representing a normalized image,
    b) generating an image vector from the normalized matrix,
    c) multiplying the image vector with a hollow matrix using a matrix-vector multiplication to generate a vector of characteristics, the hollow matrix being generated from a Gabor function which is a sine wave multiplied by a Gaussian function and the Gabor function being a function of at least one variable indicating a position in the normalized matrix and of a set of parameters comprising a parameter relating to the direction of the sine wave, a parameter relating to a center of the function of Gabor and a parameter relating to a wavelength of the sine wave,
    d) create with the vector of characteristics a probability density for a predetermined list of models,
    BE2015 / 5241
    e) select the model with the highest probability density as the best model, and
    f) classify the best model as the shape of the input image, in which there are at least two centers of the Gabor function, and in which the wavelength takes at least two values, with a first value of wavelength less than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first value of wavelength is less than a second value of wavelength and greater than or substantially equal to half of the second wavelength value.
    It has been found that by a combination of these steps, a method of pattern recognition is obtained quickly and accurately. In particular, because the characteristic vector generated in step c) is large enough to make the recognition precise and is small enough to make the calculations of the pattern recognition method fast. In addition, the choice of two wavelengths the shortest of which is greater than or equal to half of the longest and less than or equal to the distance between two centers of the Gabor function is an advantageous compromise for maintaining the number of wavelengths (and therefore the size of the feature vector) at a low level and still make the feature vector slightly redundant. This redundancy of the characteristic vector results in that a character can still be recognized even if the value of a given element of the characteristic vector is corrupted.
    In one embodiment of the invention, the models are characterized by a covariance matrix and a vector of means, the probability density being calculated by the formula:
    BE2015 / 5241
    ( r - μ) * Σ ( r - μ) where the symbol r represents the vector of characteristics, the symbol Σ represents the covariance matrix, the symbol μ represents the vector of means and k is equal to the number of elements of the vector of characteristics.
    In another embodiment, the covariance matrix is approximated. In one embodiment, the traces of the covariance matrices of all the models are equal.
    In one embodiment, all of the non-diagonal elements of the covariance matrix are set to zero. It has been found that the characterization of the models by such a covariance matrix and such a vector of means and the calculation of the probability density by this formula, in combination with the method described above, makes the recognition precise and particularly rapid.
    In an embodiment according to the present invention, the parameter relating to the direction of the sine wave is such that the angles between the possible directions of the sine wave are substantially equal.
    In an embodiment according to the invention, the parameter relating to the direction of the sine wave is an angle. In another embodiment, the sum of the highest value of the angle and the lowest value of the angle is equal to n radians. In one embodiment, the angle takes six values.
    This choice of values for the parameter relating to the direction of the sine wave gives a good compromise for obtaining sufficient angular sampling of the normalized image and for maintaining the size of the
    BE2015 / 5241 sufficiently small characteristic vector to obtain rapid process calculations.
    In another embodiment according to the invention, the Gabor function is a function of at least two parameters relating to a center of the Gabor function. In one embodiment, the parameters relating to a center of the Gabor function are such that the centers of the Gabor function are regularly spaced.
    In one embodiment of the invention, a parameter relating to a center of the Gabor function can correspond to a vertical direction in the normalized matrix and a parameter relating to a center of the Gabor function can correspond to a horizontal direction in the normalized matrix. In one embodiment, the distance from a first center of the Gabor function to an edge of the normalized matrix can be substantially equal to half the distance between two adjacent centers of the Gabor function.
    This choice of the centers of the Gabor function gives a good sampling of the normalized image and keeps the size of the vector of characteristics small enough to have rapid calculations of the process.
    In another embodiment according to the invention, the Gabor function comprises a parameter relating to the width of the Gaussian function, which can be the standard deviation of the Gaussian function. In one embodiment, the standard deviation of the Gaussian function is less than the distance between two adjacent centers of the Gabor function and greater than half the distance between two adjacent centers of the Gabor function. This choice of the width of the Gaussian function gives a good sampling of the normalized image and keeps the size of the vector of characteristics sufficiently small to obtain rapid calculations of the process.
    BE2015 / 5241
    In an embodiment according to the invention, the shape is a two-color shape, a character, an Asian character, a group of characters, a logo, an image, a drawing, a sound sequence, a film sequence or a shape. three-dimensional.
    In another embodiment according to the invention, the normalized matrix represents a normalized image with each element of the normalized matrix corresponding to a location on the normalized image. The normalized image includes the shape to be identified and the normalized matrix is the mathematical object that represents the normalized image. The normalized matrix can be processed by mathematical methods in order to extract the characteristics of the normalized image which allow the identification of the shape.
    In an embodiment according to the invention, the step of normalizing the input image into a standardized matrix comprises scaling, thresholding, smoothing, interpolation and filtering, and normalized image which corresponds to the normalized matrix has a predetermined format. The step of normalizing the input image into a standardized matrix converts the input image with the shape to be identified into a usable format. This usable format is a matrix with specific characteristics. These specific characteristics can be that the normalized matrix is binary, that in the normalized image, the standard deviation of the distance from the center of the value representing the pixels of a given color is constant.
    In embodiments according to the invention, the elements of the image vector are equal to the elements of the normalized matrix. In the step of transforming the normalized matrix into an image vector, the elements of the image vector are typically equal to the elements of the normalized matrix. Since in one embodiment the present invention, the elements of the normalized matrix are numbers
    BE2015 / 5241 binary, the elements of the image vector can be binary numbers.
    In embodiments according to the invention, the information about elements of the normalized matrix is redundant in the characteristic vector. This redundancy improves the accuracy of the identification of the shape and can be obtained by the choice of parameters, and in particular the fact that, in one embodiment of the present invention, the shorter wavelength is shorter. that the distance between two adjacent centers and the longer wavelength is longer than the distance between two adjacent centers.
    In an embodiment according to the invention, the vector of characteristics is approximated. This can increase the speed of the shape identification process calculations. The important point is the accuracy of the shape identification, the accuracy of the feature vector is of less importance.
    In an embodiment according to the invention, the normalized matrix is a 64 × 64 matrix, the image vector has 4096 elements, the hollow matrix is a 300 × 4096 matrix, the covariance matrix is a 300 × 300 matrix, the vector of means at 300 elements, the characteristic vector has 300 elements, the angle relative to the direction of the sine wave takes the values 0, 0.523598, 1.0472, 1.5708, 2.09439 and
    2.61799 radians, the centers of Gabor's function are located at positions (6,6), (6,18), (6,30), (6,42), (6,54), (18,6 ), (18,18), (18,30), (18,42), (18,54), (30,6), (30,18), (30,30), (30,42), (30.54), (42.6), (42.18), (42.30), (42.42), (42.54), (54.6), (54.18), (54 , 30), (54,42), (54,54), the first wavelength value is equal to 11, the second wavelength value is equal to 22 and the standard deviation of the gaussian function is equal to 8. It has been found that this embodiment gives very good accuracy and the corresponding calculations are rapid.
    BE2015 / 5241
    In another embodiment of the present invention, a computer program product includes a computer-readable medium in which control logic is stored to cause a computer device to identify a shape in an input image. The control logic includes:
    a) first means of computer-readable program code for normalizing the input image into a standardized matrix representing a normalized image,
    b) second means of computer readable program code for generating an image vector from the standardized matrix,
    c) third means of computer readable program code for multiplying the image vector with a sparse matrix using matrix-vector multiplication to generate a feature vector, the sparse matrix being generated from a Gabor function which is a sine wave multiplied by a Gaussian function and the Gabor function being a function of at least one variable indicating a position in the normalized matrix and of a set of parameters comprising a parameter relating to the direction of the sine wave, a parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave,
    d) fourth means of computer-readable program code for creating with the feature vector a probability density for a predetermined list of models,
    e) fifth means of computer readable program code for selecting the model with the highest probability density as the best model, and
    BE2015 / 5241
    f) sixth means of computer readable program code for classifying the best model as the shape of the input image, in which there are at least two centers of the Gabor function, and in which the length of wave takes at least two values, with a first value of wavelength less than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first value of wavelength is less than a second value of length wavelength and greater than or substantially equal to half the second wavelength value.
    In an embodiment of the present invention, the method for identifying a shape in an input image comprises the steps of:
    a) normalizing the input image into a standardized matrix representing a normalized image,
    b) create a vector of characteristics from the normalized matrix with a Gabor function, the Gabor function being a sine wave multiplied by a Gaussian function and dependent on at least one variable relating to a position on the normalized image and a set of parameters comprising a parameter relating to the direction of the sine wave, at least one parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave,
    c) generating from the vector of characteristics and from a predetermined list of models 108 a probability density of each model,
    d) identify the model with the highest probability density as the shape in an input image,
    BE2015 / 5241 in which there are at least two centers of the Gabor function, and the wavelength takes values in a first set of at least one value and in a second set of at least one value with
    at. the first set of at least one value less than or substantially equal to the distance between two adjacent centers of the Gabor function, and
    b. the first set of at least one value greater than or substantially equal to half the values in the second set of at least one value.
    It has been found that by a combination of these steps, a method of pattern recognition is obtained quickly and accurately. In particular, firstly, the characteristic vector generated in step c) is large enough to make the recognition precise and is small enough to make the calculations of the pattern recognition method fast. In addition, the choice of two wavelengths the shorter of which is greater than or equal to half of the longest and less than or equal to the distance between two centers of the Gabor function is a good compromise for maintaining the number of lengths wave (and therefore the size of the feature vector) at a low level and still make the feature vector slightly redundant.
    In one embodiment of the invention, the models are characterized by a covariance matrix and a vector of means. The probability density can be calculated by the formula
    ν (2'π) ^ ΪΣΪ
    BE2015 / 5241 where the symbol r represents the vector of characteristics, the symbol Σ represents the covariance matrix, the symbol μ represents the vector of means and k is equal to the number of elements of the vector of characteristics. In embodiments, all non-diagonal elements of the covariance matrix can be set to zero, the covariance matrix can be approximated, the traces of the covariance matrices of all models can be equal. It has been found that the characterization of the models by such a covariance matrix and such a vector of means and the calculation of the probability density by this formula, in combination with the method described above, makes the recognition precise and particularly rapid.
    In embodiments according to the invention, the parameter relating to the direction of the sine wave is such that the angles between the possible directions of the sine wave are substantially equal.
    In an embodiment according to the invention, the parameter relating to the direction of the sine wave is an angle and the sum of its highest value and its lowest value is equal to n radians. This angle can take six values. This choice of values for the parameter relating to the direction of the sine wave gives a good compromise for obtaining sufficient angular sampling of the normalized image and for keeping the size of the vector of characteristics small enough to obtain rapid calculations of the process.
    In embodiments according to the invention, at least two parameters relate to a center of the Gabor function and can be such that the centers of the Gabor function are regularly spaced. A parameter relating to a center of the Gabor function can correspond to a vertical direction in the normalized matrix and a parameter relating to a center of the Gabor function can correspond to a horizontal direction in the normalized matrix. The distance of one
    BE2015 / 5241 first center of the Gabor function at an edge of the normalized matrix can be substantially equal to half the distance between two adjacent centers of the Gabor function. This choice of the centers of the Gabor function gives a good sampling of the normalized image and keeps the size of the vector of characteristics small enough to obtain rapid calculations of the process.
    In embodiments according to the invention, a parameter relates to the width of the Gaussian function, which can be the standard deviation of the Gaussian function. This standard deviation of the Gaussian function is less than the distance between two adjacent centers of the Gabor function and may be greater than half the distance between two adjacent centers of the Gabor function. This choice of the width of the Gaussian function gives a good sampling of the normalized image, makes the vector of characteristics slightly redundant and keeps the size of the vector of characteristics sufficiently small to obtain rapid calculations of the process.
    In embodiments according to the invention, the shape is a two-color shape, a character, an Asian character, a group of characters, a logo, an image, a drawing, a sound sequence, a film sequence or a shape. three-dimensional.
    In embodiments according to the invention, the normalized matrix represents a normalized image with each element of the normalized matrix corresponding to a location on the normalized image. The normalized image includes the shape to be identified and the normalized matrix is the mathematical object that represents the normalized image. The normalized matrix can be processed by mathematical methods in order to extract the characteristics of the normalized image which allow the identification of the shape.
    BE2015 / 5241
    In embodiments according to the invention, the step of normalizing the input image into a standardized matrix comprises scaling, thresholding, smoothing, interpolation and filtering, and normalized image which corresponds to the normalized matrix has a predetermined format. The step of normalizing the input image into a standardized matrix converts the input image with the shape to be identified into a usable format. This usable format is a matrix with specific characteristics. These specific characteristics can be that the normalized matrix is binary, that in the normalized image, the standard deviation of the distance from the center of the value representing the pixels of a given color is constant.
    In embodiments according to the invention, the elements of an image vector are equal to the elements of the normalized matrix. In the step of transforming the normalized matrix into an image vector, the elements of the image vector are typically established equal to the elements of the normalized matrix. Since in one embodiment of the present invention, the elements of the normalized matrix are binary numbers, the elements of the image vector can be binary numbers.
    In embodiments according to the invention, the information about elements of the normalized matrix is redundant in the characteristic vector. This redundancy improves the accuracy of the identification of the shape and can be obtained by the choice of parameters, and in particular the fact that, in one embodiment of the present invention, the standard deviation of the Gaussian function of the Gabor function is less than the distance between two adjacent centers but greater than half the distance between two adjacent centers.
    In embodiments according to the invention, the vector of characteristics is approximated. This can increase the speed of calculations
    BE2015 / 5241 of the form identification process. The important point is the accuracy of the shape identification, the accuracy of the feature vector is of less importance.
    In one embodiment according to the invention, the normalized matrix is a 64 × 64 matrix, the image vector has 4096 elements, the hollow matrix used in the step of creating the characteristic vector is a 300 × 4096 matrix, the matrix of covariance is a matrix of 300x300, the vector of means has 300 elements, the vector of characteristics has 300 elements, the angle relative to the direction of the sine wave takes the values 0, 0.523598, 1.0472, 1 , 5708, 2.09439 and
    2.61799 radians, the centers of Gabor's function are located at positions (6,6), (6,18), (6,30), (6,42), (6,54), (18,6 ), (18,18), (18,30), (18,42), (18,54), (30,6), (30,18), (30,30), (30,42), (30.54), (42.6), (42.18), (42.30), (42.42), (42.54), (54.6), (54.18), (54 , 30), (54,42), (54,54), the first wavelength value is equal to 11, the second wavelength value is equal to 22 and the standard deviation of the gaussian function is equal to 8. It has been found that this embodiment gives very good accuracy and the corresponding calculations are rapid.
    In another embodiment of the present invention, a computer program product includes a computer-readable medium in which is stored control logic for causing a computer device to identify a shape in an input image. The control logic includes:
    a) first means of computer-readable program code for normalizing the input image into a standardized matrix representing a normalized image,
    b) second means of computer-readable program code for creating a vector of characteristics from the normalized matrix with a Gabor function, the function of
    BE2015 / 5241
    Gabor being a sine wave multiplied by a Gaussian function and dependent on at least one variable relating to a position on the normalized image and on a set of parameters comprising a parameter relating to the direction of the sine wave, at least one parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave,
    c) third means of computer readable program code for generating from the vector of characteristics and from a predetermined list of models a probability density of each model to identify a better model such as the shape in an input image,
    d) fourth means of computer readable program code for identifying the model with the highest probability density as the shape in an input image, in which there are at least two centers of the Gabor function, and the wavelength takes values in a first set of at least one value and in a second set of at least one value with
    at. the first set of at least one value less than or substantially equal to the distance between two adjacent centers of the Gabor function, and
    b. the first set of at least one value greater than or substantially equal to half the values in the second set of at least one value.
    BRIEF DESCRIPTION OF THE DRAWINGS
    BE2015 / 5241
    For a better understanding of the present invention, reference will now be made, by way of example, to the accompanying drawings in which:
    Figure 1 shows a flow diagram of a character recognition process according to the invention.
    FIG. 2 shows a schematic illustration of a normalization step in an optical character recognition process according to the invention.
    FIG. 3 shows a flow diagram of an extraction step in an optical character recognition process according to the invention.
    Figure 4 shows a flowchart which describes how the elements of the hollow matrix are generated in an optical character recognition process according to the invention.
    FIG. 5 a shows an illustration of a matrix multiplication between a hollow matrix and an image vector used in an optical character recognition process according to the invention.
    FIG. 5 b shows an illustration of a threshold matrix used in an optical character recognition process according to the invention.
    FIG. 6 shows a flow diagram of a classification step in an optical character recognition process according to the invention.
    DESCRIPTION OF THE INVENTION
    The present invention will be described in relation to particular embodiments and with reference to certain drawings, but the invention is however not limited thereto. The drawings described are only schematic
    BE2015 / 5241 and are not limiting. In the drawings, the size of some of the elements may be exaggerated and not drawn to scale for illustrative purposes.
    In addition, the terms first, second, third and the like in the description and in the claims are used to distinguish between similar elements and not necessarily to describe a sequential or chronological order. The terms are interchangeable under the appropriate circumstances and the embodiments of the invention can function in sequences other than those described or illustrated in this document.
    In addition, the various embodiments, although qualified as preferred, should be interpreted as exemplary ways in which the invention can be implemented rather than as limiting the scope of the invention.
    The term comprising, used in the claims, should not be interpreted as being limited to the means or steps listed below; it does not exclude other elements or stages. It must be interpreted as specifying the presence of the elements, whole numbers, stages or components mentioned to which reference is made, but does not exclude the presence or addition of one or more other elements, whole numbers, stages or components or groups. of these. Therefore, the scope of the expression a device comprising A and B should not be limited to devices comprising only the components A and B, on the contrary, as regards the present invention, the only listed components of the device are A and B, and the claim should also be interpreted to include equivalents of these components.
    Binary numbers, vectors and matrices are supposed to be written here with 0 and 1 but it is clear to a person skilled in the art that they could be written as true and false, black and white or any other means to indicate a binary state.
    BE2015 / 5241
    In one embodiment of the present invention, binary images are processed. Binary images are digital images with only two possible colors for each pixel. These two colors, generally black and white, can be represented as true and false values or values 1 and 0. A representation with 1 and 0 is especially useful for performing mathematical image processing. Binary image processing often includes filtering steps in order, for example, to improve certain characteristics of the image, or to perform morphological operations on the image. Filters are generally described mathematically by matrices and the application of a filter on a binary image is described by the multiplication of the filter matrix and the binary image matrix. This type of operation can be used, for example in optical character recognition, as a step in image processing to extract the characteristics of the image in order to recognize an optical character.
    Optical character recognition systems convert the image of text into machine readable code using a character recognition process. In an OCR system, images of what could be characters are isolated and a character recognition method is used to identify the character.
    An embodiment of the present invention relates to optical character recognition starting from an input image representing a character or another shape. In a preferred embodiment of the present invention, optical character recognition starts from an input image representing an Asian character. The input image is, in one embodiment of the invention, a two-color image. In a preferred embodiment of the present invention, the input image is a black and white image. In one embodiment of the present invention, the input image is a two-dimensional image. In
    BE2015 / 5241 an embodiment of the present invention, the input image represents a shape comprising a character, a logo, an image or a design, to be recognized by the recognition system. In another embodiment of the present invention, the shape is a sound sequence, a movie sequence or a three-dimensional shape, to be recognized by the recognition system.
    An optical character recognition process 101 according to an embodiment of the invention shown in FIG. 1 comprises:
    · A normalization step 103 which generates a normalized matrix 104 from an input image 102;
    · An extraction step 105 which generates a vector of characteristics 106 from the standardized matrix 104; and · a classification step 107 which calculates a best model 109 for the input image 102 from a series of possible models 108. The classification step 107 also returns a probability density 110 of each model, which provides a measure of the accuracy of the classification step 107.
    In the normalization step 103, the input image 102 is subdivided into pixels 201. Each pixel 201 of the input image 102 is represented by an element 202 of an intermediate matrix 203, as illustrated in the figure 2. The intermediate matrix 203 is, in one embodiment of the invention, a binary matrix. The intermediate matrix 203 is subjected to a combination of steps which may include scaling, thresholding, smoothing, interpolation, filtering, etc. The product of this combination of steps is a standardized matrix 104 which corresponds to a standardized image 206. In one embodiment of the
    BE2015 / 5241 present invention, the normalized image 206 has a predetermined format and is centered.
    Each element 207 of the standardized matrix 104 corresponds to a pixel 208 of the predetermined format in the standardized image 206.
    In one embodiment of the present invention, the normalized matrix 104 is a binary matrix which corresponds to a normalized image 206 two-color. Each element of the standardized matrix 104 is characterized by its line x 204 and its column y 205, which corresponds to a location on the standardized image 206. In one embodiment of the present invention, the standardized matrix 104 is a matrix of 64x64. In one embodiment of the present invention, in the normalized image 206, the standard deviation of the distance from the center of the value representing the pixels of a given color is constant. In a preferred embodiment of the present invention, in the normalized image 206, the standard deviation of the distance from the center of the value representing the pixels of a given color is equal to 16 pixels. In one embodiment of the present invention, the aspect ratio of aspect ratio of shape or character is preserved during the normalization step 103.
    The feature extraction step 105 which generates the feature vector 106 from the normalized matrix 104 includes a matrix-vector multiplication 304. This can be explained in detail using FIG. 3. The normalized matrix 104 of dimensions AxB is transformed into an image vector 301 of length A * B. During this transformation, each element of the image vector 301 is set equal to an element of the normalized matrix 104 so that all the elements of the normalized matrix 104 are copied only once in the image vector 301. In a mode As an embodiment of the present invention, the image vector 301 is a binary vector. The image vector 301 contains the information of the normalized image 206. The location of an element of the
    BE2015 / 5241 line x 204 and column y 205 in the normalized matrix 104, i.e. which corresponds to a location in the normalized image 206, also corresponds to a specific value of the index j 302 which indicates the j th element of the image vector 301. In an embodiment of the present invention, the image vector 301 has 4096 elements and the index j can take any integer values between 1 and 4096. This corresponds to a normalized matrix 104 of 64x64 (64 * 64 = 4096).
    In one embodiment of the present invention, the matrix-vector multiplication 304 is approximate and the characteristic vector 106 is an approximation of the exact mathematical result of the matrix multiplication between a hollow matrix 303 and the image vector 301. A subscript i 401 is used to specify the i th element of the feature vector 106. The hollow adjective indicates that the matrix is populated mainly with zeros in an embodiment of the present invention.
    Figure 4 describes the generation, with a Gabor function 404, of an element 406 located at line i 401 and column j 302 of the hollow matrix 303. All the elements 406 of the hollow matrix 303 are generated in the same way way. The Gabor 404 function is a plane sine wave multiplied by a Gaussian function. The Gabor function 404 has parameters 402, which correspond to the index i, and variables x 204 and y 205, which correspond to the index j, as inputs.
    The line index i 401 of the hollow matrix element 406 to be calculated specifies the values taken by the parameters 402 used in the Gabor function 404. In one embodiment of the present invention, the parameters 402 are represented by the symbols ai, σ ,, λί, Cx and Cy:
    · A i is an angle relative to the direction of the plane sine wave of the Gabor function 404;
    BE2015 / 5241 · o, is the standard deviation of the Gaussian function from the Gabor 404 function;
    · Λί is the wavelength of the plane sine wave of the Gabor function 404;
    · Cxi is the center of the function of Gabor 404 on the normalized image 206, in the vertical direction; and · Cyi is the center of the Gabor function 404 on the normalized image 206, in the horizontal direction.
    In one embodiment of the invention, the parameters are chosen as follows:
    · Cxi values are regularly spaced.
    · If the distance between two values Cxi is called dCx, the first Cxi, Cx1 is equal to dCx / 2.
    · Cyi values are regularly spaced.
    · If the distance between two Cyi values is called dCy, the first Cyi, Cy1 is equal to dCy / 2.
    · The angles a i are regularly spaced.
    · The sum of the highest value of a i and the lowest value of a i is equal to n radians.
    · The values of o, are less than dCx.
    · The values of o, are greater than dCx / 2.
    · The values of o, are less than dCy.
    · The values of oi are greater than dCy / 2.
    · In a first set of values of at least one value, the values of λ i are less than dCx and dCy.
    · The values of λ i in a second set of at least one value are such that the values in the first set of values are greater than or substantially equal to half the values in the second set of values.
    BE2015 / 5241
    In additional embodiments of the invention, the parameters meet one or more of the following conditions: · dCx and dCy are equal to each other.
    · Cxi takes five values.
    · Cyi takes five values.
    · A i takes six values.
    · Ai takes the values 0, 0.523598, 1.0472, 1.5708, 2.09439 and 2.61799 radians.
    · Σ, takes a value.
    · Λ, takes two values: λι and λ2.
    · Λ1 is less than dCx.
    · Λ1 is less than dCy.
    · Λ2 is greater than λ1 and less than or equal to 2 * λ1.
    In a preferred embodiment of a standard matrix 104 of 64 × 64, the positions Cxi are 6, 18, 30, 42 and 54, the positions Cyi are 6, 18, 30, 42 and 54, the standard deviation σ, is 8, the wavelengths λ, are 11 and 22. The values of the parameters in an embodiment of the present invention are given in Table 1.
    The number of sets of values for parameters 402 is equal to (number of values of Cxi * number of values of Cyi * number of values of a * number of values of σ i * number of values of λ,). In a preferred embodiment, the number of parameter sets is equal to 300 = 5 * 5 * 6 * 1 * 2 and the row index i 401 takes integer values from 1 to 300.
    The column index j 302 of the hollow matrix element 406 to be calculated specifies the values of the variables x 204 and y 205 used by the Gabor function 404.
    The function of Gabor 404 is expressed by:
    BE2015 / 5241 g [i, (x, y)] = π σ, 2
    [cos (a,) (x - Cx,) + sin ( aj) ( y - Cy, )] - (x - Cx,) 2 + (y - Cy,) 2
    2σ, 2 the product of the Gabor function 404 calculated · from a given set of values of parameters 402 which corresponds to the index i 401, · at a given location in line x 204 and column y 205 of the normalized image 206 which corresponds to the index j 302 is the element 406 of the line i 401 and of the column j 302 in the hollow matrix 303. The number of columns of the hollow matrix 303 is equal to the number d Elements of the image vector 301. In one embodiment of the present invention, the hollow matrix 303 is a 300x4096 matrix.
    In one embodiment of the present invention, the element (i, j) 406 of the hollow matrix is given by (V ~ 1 2 π
    Mij = ----- 2 exp --- r ---- [ cos ( ai ) ( (j - 1)% 64 - Cxi ) π σ, 2 I λ, + sin ( a, ) ( (j - 1) // 64 - Cy, )] ((j - 1)% 64 - Cxi) 2 + ((j - 1) // 64 - Cy,) 2
    2σ, 2
    The% symbol represents the modulo operation and takes precedence over the * and / operations, and the // symbol gives the integer part of the result of a division between 15 whole numbers and takes precedence over the * and / operations. The parameters have the following values
    BE2015 / 5241 αι = (i - 1)% nb a * nb 0
    2 3 ^ 9 % n7:
    Φ1 = ------------- maxWavelength
    Cyî = stepSize // 2 + stepSize * (- ^ --- ~ 9% -bSteps
    Cxj = stepSize // 2 + stepSize * (n --- / --- J% nbFeatures nb 8 * nb: * nbFtGHi / nb0 = 6 nbD = 2 maxWavelength = 22 stepSize = 12 nbSteps = 5 nbFeatures = 300 σ = 8.
    nbSteps is such that Cxi and Cyi are less than or equal to 64.
    A matrix-vector multiplication 304 is performed to multiply the hollow matrix 303 and the image vector 301, the hollow matrix 303 being the first factor of the multiplication and the image vector 301 being the second factor of the multiplication as illustrated in FIG. 5. The vector resulting from the multiplication of the hollow matrix 303 and of the image vector 301 is the vector of characteristics 106. The number of elements of the vector of characteristics 106 is equal to the number of lines
    BE2015 / 5241 of the hollow matrix 303. In a preferred embodiment of the present invention, the number of elements of the characteristic vector 106 is equal to 300.
    In one embodiment of the present invention, the feature vector 106 contains specific information about the input image 102, this specific information regarding the image features important for pattern recognition. The choice of parameters, and in particular the fact that, in one embodiment of the present invention, the standard deviation of the Gaussian function from the Gabor function is less than the distance between two adjacent centers but greater than half of the distance between two adjacent centers makes the information contained in the characteristic vector 106 slightly redundant. The redundancy of the information in the characteristic vector 106 increases the accuracy of the classification step 107.
    The matrix multiplication between the hollow matrix 303 and the image vector 301 resulting in the characteristic vector 106 is shown in Figure 5a. The elements of the hollow matrix 303 are called Mij. i is the index which gives the line number and takes all the integer values between 1 and m. j is the index which gives the column number and takes all the integer values between 1 and n. Image vector 301 has a column of n elements called vj. The feature vector 106 has a column of m elements called ri. The multiplication of the matrix is such that the elements ri of the vector of characteristics 106 are calculated as ri = SjO = 5 Mij vj (Equation 1)
    Certain terms can be overlooked in the sum of Equation 1. For example, the terms Mijvj where vj is zero is
    BE2015 / 5241 also equal to 0. In addition, in the case where vj is equal to 1, and where the element Mij of the hollow matrix 303 is small, the term Mijvj can also be neglected. To control small, a threshold matrix 501 with Tij elements, shown in Figure 5b, is used in an embodiment of the present invention. In one embodiment of the present invention, a term Mij vj can be neglected if Mij is less than Tij. In a further embodiment of the present invention, all the elements Tij of the threshold matrix 501 have the same value. Since the Gabor function 404 is a plane sine wave multiplied by a Gaussian function, many elements of the hollow matrix 303 are very small.
    The classification step 107 of the OCR process 101 can be described with the aid of FIG. 6. In an embodiment of the present invention, the classification step 107 is a variation of the closest classifier method neighbors using the weighted Euclidean distance where the weights are different for each class. The classification step 107 uses the characteristic vector 106 and the models 108 as inputs. In one embodiment of the present invention, the models 108 correspond to characters, groups of characters or characters in a given font family. In one embodiment of the present invention, the models 108 correspond to Asian characters, groups of Asian characters or Asian characters in a given font family. In one embodiment of the present invention, the models 108 correspond to sound sequences, film sequences or three-dimensional shapes.
    In an embodiment of the present invention, a model 108 is defined by a covariance matrix Σ and a vector of means μ. In one embodiment of the present invention, all
    BE2015 / 5241 non-diagonal elements of Σ are set to zero. In one embodiment of the present invention, the covariance matrices Σ are multiplied by a constant (different constant for each model) so that the traces of the covariance matrices Σ of all the models are equal. In one embodiment of the present invention, the covariance matrix is approximated. In an embodiment of the present invention, Σ is a matrix of 300x300 and μ a vector of 300 elements.
    To select the model which best corresponds to the input image 102 corresponding to the characteristic vector 106, a probability density 110 is calculated, for each model 108, as
    where the symbol r represents the vector of characteristics 106.
    The symbol | Σ | represents the determinant of the matrix Σ and the t in (r - μ) 1 indicates the transposition of the vector (r - μ). k is equal to the number of elements of the characteristic vector 106. In an embodiment of the present invention, k is equal to 300. The product (r μ) 1 Σ (r - μ) is a matrix multiplication according to the usual mathematical conventions.
    Once the 601 probability density of each model
    108 is calculated in a calculation step 601, the best model 109 is selected in a selection step 602. The best model 109 is the model with the highest probability density 110. In one embodiment of the present invention, the The classification step 107 makes the best model 109 and the probability density 110 of each model, to provide a measure of the accuracy of the classification step. In an alternative embodiment, the classification step 107 returns only the best model 109. In an alternative embodiment, the step of
    BE2015 / 5241 classification 107 only gives the probability density of each model 110.
    An embodiment of the present invention comprises the combination of
    the step for extracting characteristics 105 with the parameters 402 as described above, and
    - the classification step 107 based on weighted Euclidean distances with all the non-diagonal elements of the covariance matrix Σ set to zero and the traces of the covariance matrices Σ of all equal models.
    Such an OCR system allows image reconstruction, can be calculated efficiently and makes the accuracy of OCR extremely high.
    BE2015 / 5241
    Table 1
    index xy Ζ / [γ ] γ Cx i Cyi 1 8 0.0454545 0 6 6 2 8 0.0454545 0.523598 6 6 3 8 0.0454545 1.0472 6 6 4 8 0.0454545 1.5708 6 6 5 8 0.0454545 2.09439 6 6 6 8 0.0454545 2.61799 6 6 7 8 0.0909091 0 6 6 8 8 0.0909091 0.523598 6 6 9 8 0.0909091 1.0472 6 6 10 8 0.0909091 1.5708 6 6 11 8 0.0909091 2.09439 6 6 12 8 0.0909091 2.61799 6 6 13 8 0.0454545 0 6 18 14 8 0.0454545 0.523598 6 18 15 8 0.0454545 1.0472 6 18 16 8 0.0454545 1.5708 6 18 17 8 0.0454545 2.09439 6 18 18 8 0.0454545 2.61799 6 18 19 8 0.0909091 0 6 18 20 8 0.0909091 0.523598 6 18 21 8 0.0909091 1.0472 6 18 22 8 0.0909091 1.5708 6 18 23 8 0.0909091 2.09439 6 18 24 8 0.0909091 2.61799 6 18 25 8 0.0454545 0 6 30 26 8 0.0454545 0.523598 6 30 27 8 0.0454545 1.0472 6 30 28 8 0.0454545 1.5708 6 30 29 8 0.0454545 2.09439 6 30 30 8 0.0454545 2.61799 6 30 31 8 0.0909091 0 6 30 32 8 0.0909091 0.523598 6 30 33 8 0.0909091 1.0472 6 30 34 8 0.0909091 1.5708 6 30
    BE2015 / 5241
    35 8 0.0909091 2.09439 6 30 36 8 0.0909091 2.61799 6 30 37 8 0.0454545 0 6 42 38 8 0.0454545 0.523598 6 42 39 8 0.0454545 1.0472 6 42 40 8 0.0454545 1.5708 6 42 41 8 0.0454545 2.09439 6 42 42 8 0.0454545 2.61799 6 42 43 8 0.0909091 0 6 42 44 8 0.0909091 0.523598 6 42 45 8 0.0909091 1.0472 6 42 46 8 0.0909091 1.5708 6 42 47 8 0.0909091 2.09439 6 42 48 8 0.0909091 2.61799 6 42 49 8 0.0454545 0 6 54 50 8 0.0454545 0.523598 6 54 51 8 0.0454545 1.0472 6 54 52 8 0.0454545 1.5708 6 54 53 8 0.0454545 2.09439 6 54 54 8 0.0454545 2.61799 6 54 55 8 0.0909091 0 6 54 56 8 0.0909091 0.523598 6 54 57 8 0.0909091 1.0472 6 54 58 8 0.0909091 1.5708 6 54 59 8 0.0909091 2.09439 6 54 60 8 0.0909091 2.61799 6 54 61 8 0.0454545 0 18 6 62 8 0.0454545 0.523598 18 6 63 8 0.0454545 1.0472 18 6 64 8 0.0454545 1.5708 18 6 65 8 0.0454545 2.09439 18 6 66 8 0.0454545 2.61799 18 6 67 8 0.0909091 0 18 6 68 8 0.0909091 0.523598 18 6 69 8 0.0909091 1.0472 18 6 70 8 0.0909091 1.5708 18 6 71 8 0.0909091 2.09439 18 6 72 8 0.0909091 2.61799 18 6
    BE2015 / 5241
    73 8 0.0454545 0 18 18 74 8 0.0454545 0.523598 18 18 75 8 0.0454545 1.0472 18 18 76 8 0.0454545 1.5708 18 18 ΊΊ 8 0.0454545 2.09439 18 18 78 8 0.0454545 2.61799 18 18 79 8 0.0909091 0 18 18 80 8 0.0909091 0.523598 18 18 81 8 0.0909091 1.0472 18 18 82 8 0.0909091 1.5708 18 18 83 8 0.0909091 2.09439 18 18 84 8 0.0909091 2.61799 18 18 85 8 0.0454545 0 18 30 86 8 0.0454545 0.523598 18 30 87 8 0.0454545 1.0472 18 30 88 8 0.0454545 1.5708 18 30 89 8 0.0454545 2.09439 18 30 90 8 0.0454545 2.61799 18 30 91 8 0.0909091 0 18 30 92 8 0.0909091 0.523598 18 30 93 8 0.0909091 1.0472 18 30 94 8 0.0909091 1.5708 18 30 95 8 0.0909091 2.09439 18 30 96 8 0.0909091 2.61799 18 30 97 8 0.0454545 0 18 42 98 8 0.0454545 0.523598 18 42 99 8 0.0454545 1.0472 18 42 100 8 0.0454545 1.5708 18 42 101 8 0.0454545 2.09439 18 42 102 8 0.0454545 2.61799 18 42 103 8 0.0909091 0 18 42 104 8 0.0909091 0.523598 18 42 105 8 0.0909091 1.0472 18 42 106 8 0.0909091 1.5708 18 42 107 8 0.0909091 2.09439 18 42 108 8 0.0909091 2.61799 18 42 109 8 0.0454545 0 18 54 110 8 0.0454545 0.523598 18 54
    BE2015 / 5241
    111 8 0.0454545 1.0472 18 54 112 8 0.0454545 1.5708 18 54 113 8 0.0454545 2.09439 18 54 114 8 0.0454545 2.61799 18 54 115 8 0.0909091 0 18 54 116 8 0.0909091 0.523598 18 54 117 8 0.0909091 1.0472 18 54 118 8 0.0909091 1.5708 18 54 119 8 0.0909091 2.09439 18 54 120 8 0.0909091 2.61799 18 54 121 8 0.0454545 0 30 6 122 8 0.0454545 0.523598 30 6 123 8 0.0454545 1.0472 30 6 124 8 0.0454545 1.5708 30 6 125 8 0.0454545 2.09439 30 6 126 8 0.0454545 2.61799 30 6 127 8 0.0909091 0 30 6 128 8 0.0909091 0.523598 30 6 129 8 0.0909091 1.0472 30 6 130 8 0.0909091 1.5708 30 6 131 8 0.0909091 2.09439 30 6 132 8 0.0909091 2.61799 30 6 133 8 0.0454545 0 30 18 134 8 0.0454545 0.523598 30 18 135 8 0.0454545 1.0472 30 18 136 8 0.0454545 1.5708 30 18 137 8 0.0454545 2.09439 30 18 138 8 0.0454545 2.61799 30 18 139 8 0.0909091 0 30 18 140 8 0.0909091 0.523598 30 18 141 8 0.0909091 1.0472 30 18 142 8 0.0909091 1.5708 30 18 143 8 0.0909091 2.09439 30 18 144 8 0.0909091 2.61799 30 18 145 8 0.0454545 0 30 30 146 8 0.0454545 0.523598 30 30 147 8 0.0454545 1.0472 30 30 148 8 0.0454545 1.5708 30 30
    BE2015 / 5241
    149 8 0.0454545 2.09439 30 30 150 8 0.0454545 2.61799 30 30 151 8 0.0909091 0 30 30 152 8 0.0909091 0.523598 30 30 153 8 0.0909091 1.0472 30 30 154 8 0.0909091 1.5708 30 30 155 8 0.0909091 2.09439 30 30 156 8 0.0909091 2.61799 30 30 157 8 0.0454545 0 30 42 158 8 0.0454545 0.523598 30 42 159 8 0.0454545 1.0472 30 42 160 8 0.0454545 1.5708 30 42 161 8 0.0454545 2.09439 30 42 162 8 0.0454545 2.61799 30 42 163 8 0.0909091 0 30 42 164 8 0.0909091 0.523598 30 42 165 8 0.0909091 1.0472 30 42 166 8 0.0909091 1.5708 30 42 167 8 0.0909091 2.09439 30 42 168 8 0.0909091 2.61799 30 42 169 8 0.0454545 0 30 54 170 8 0.0454545 0.523598 30 54 171 8 0.0454545 1.0472 30 54 172 8 0.0454545 1.5708 30 54 173 8 0.0454545 2.09439 30 54 174 8 0.0454545 2.61799 30 54 175 8 0.0909091 0 30 54 176 8 0.0909091 0.523598 30 54 177 8 0.0909091 1.0472 30 54 178 8 0.0909091 1.5708 30 54 179 8 0.0909091 2.09439 30 54 180 8 0.0909091 2.61799 30 54 181 8 0.0454545 0 42 6 182 8 0.0454545 0.523598 42 6 183 8 0.0454545 1.0472 42 6 184 8 0.0454545 1.5708 42 6 185 8 0.0454545 2.09439 42 6 186 8 0.0454545 2.61799 42 6
    BE2015 / 5241
    187 8 0.0909091 0 42 6 188 8 0.0909091 0.523598 42 6 189 8 0.0909091 1.0472 42 6 190 8 0.0909091 1.5708 42 6 191 8 0.0909091 2.09439 42 6 192 8 0.0909091 2.61799 42 6 193 8 0.0454545 0 42 18 194 8 0.0454545 0.523598 42 18 195 8 0.0454545 1.0472 42 18 196 8 0.0454545 1.5708 42 18 197 8 0.0454545 2.09439 42 18 198 8 0.0454545 2.61799 42 18 199 8 0.0909091 0 42 18 200 8 0.0909091 0.523598 42 18 201 8 0.0909091 1.0472 42 18 202 8 0.0909091 1.5708 42 18 203 8 0.0909091 2.09439 42 18 204 8 0.0909091 2.61799 42 18 205 8 0.0454545 0 42 30 206 8 0.0454545 0.523598 42 30 207 8 0.0454545 1.0472 42 30 208 8 0.0454545 1.5708 42 30 209 8 0.0454545 2.09439 42 30 210 8 0.0454545 2.61799 42 30 211 8 0.0909091 0 42 30 212 8 0.0909091 0.523598 42 30 213 8 0.0909091 1.0472 42 30 214 8 0.0909091 1.5708 42 30 215 8 0.0909091 2.09439 42 30 216 8 0.0909091 2.61799 42 30 217 8 0.0454545 0 42 42 218 8 0.0454545 0.523598 42 42 219 8 0.0454545 1.0472 42 42 220 8 0.0454545 1.5708 42 42 221 8 0.0454545 2.09439 42 42 222 8 0.0454545 2.61799 42 42 223 8 0.0909091 0 42 42 224 8 0.0909091 0.523598 42 42
    BE2015 / 5241
    225 8 0.0909091 1.0472 42 42 226 8 0.0909091 1.5708 42 42 227 8 0.0909091 2.09439 42 42 228 8 0.0909091 2.61799 42 42 229 8 0.0454545 0 42 54 230 8 0.0454545 0.523598 42 54 231 8 0.0454545 1.0472 42 54 232 8 0.0454545 1.5708 42 54 233 8 0.0454545 2.09439 42 54 234 8 0.0454545 2.61799 42 54 235 8 0.0909091 0 42 54 236 8 0.0909091 0.523598 42 54 237 8 0.0909091 1.0472 42 54 238 8 0.0909091 1.5708 42 54 239 8 0.0909091 2.09439 42 54 240 8 0.0909091 2.61799 42 54 241 8 0.0454545 0 54 6 242 8 0.0454545 0.523598 54 6 243 8 0.0454545 1.0472 54 6 244 8 0.0454545 1.5708 54 6 245 8 0.0454545 2.09439 54 6 246 8 0.0454545 2.61799 54 6 247 8 0.0909091 0 54 6 248 8 0.0909091 0.523598 54 6 249 8 0.0909091 1.0472 54 6 250 8 0.0909091 1.5708 54 6 251 8 0.0909091 2.09439 54 6 252 8 0.0909091 2.61799 54 6 253 8 0.0454545 0 54 18 254 8 0.0454545 0.523598 54 18 255 8 0.0454545 1.0472 54 18 256 8 0.0454545 1.5708 54 18 257 8 0.0454545 2.09439 54 18 258 8 0.0454545 2.61799 54 18 259 8 0.0909091 0 54 18 260 8 0.0909091 0.523598 54 18 261 8 0.0909091 1.0472 54 18 262 8 0.0909091 1.5708 54 18
    BE2015 / 5241
    263 8 0.0909091 2.09439 54 18 264 8 0.0909091 2.61799 54 18 265 8 0.0454545 0 54 30 266 8 0.0454545 0.523598 54 30 267 8 0.0454545 1.0472 54 30 268 8 0.0454545 1.5708 54 30 269 8 0.0454545 2.09439 54 30 270 8 0.0454545 2.61799 54 30 271 8 0.0909091 0 54 30 272 8 0.0909091 0.523598 54 30 273 8 0.0909091 1.0472 54 30 274 8 0.0909091 1.5708 54 30 275 8 0.0909091 2.09439 54 30 276 8 0.0909091 2.61799 54 30 277 8 0.0454545 0 54 42 278 8 0.0454545 0.523598 54 42 279 8 0.0454545 1.0472 54 42 280 8 0.0454545 1.5708 54 42 281 8 0.0454545 2.09439 54 42 282 8 0.0454545 2.61799 54 42 283 8 0.0909091 0 54 42 284 8 0.0909091 0.523598 54 42 285 8 0.0909091 1.0472 54 42 286 8 0.0909091 1.5708 54 42 287 8 0.0909091 2.09439 54 42 288 8 0.0909091 2.61799 54 42 289 8 0.0454545 0 54 54 290 8 0.0454545 0.523598 54 54 291 8 0.0454545 1.0472 54 54 292 8 0.0454545 1.5708 54 54 293 8 0.0454545 2.09439 54 54 294 8 0.0454545 2.61799 54 54 295 8 0.0909091 0 54 54 296 8 0.0909091 0.523598 54 54 297 8 0.0909091 1.0472 54 54 298 8 0.0909091 1.5708 54 54 299 8 0.0909091 2.09439 54 54 300 8 0.0909091 2.61799 54 54
    BE2015 / 5241
    References of the figures
    Figure 1
    102 Input image
    103 Standardization
    104 Standardized matrix
    105 Feature extraction
    106 Characteristic vector
    107 Classification
    108 Models
    109 Best model
    110 Probability density of each model
    Figure 3
    104 Standardized matrix
    204 Line x
    205 column y
    301 Image vector
    302 Clue j
    303 Hollow matrix
    304 Multiplication matrix - vector
    106 Characteristic vector
    Figure 4
    401 Index i
    402 Settings
    404 Gabor function
    204 Line x
    205 column y
    406 Element (i, j) of the hollow matrix
    BE2015 / 5241
    302 Clue j
    Figure 6
    106 Characteristic vector
    107 Classification
    601 Calculation of the probability density of each model
    602 Selection of the best model
    108 Models
    109 Best model
    权利要求:
    Claims (6)
    [1]
    1. A method for identifying a shape in an input image, comprising the steps of
    a) normalizing the input image into a standardized matrix representing a normalized image,
    b) generating an image vector from the normalized matrix,
    c) multiplying the image vector with a hollow matrix using a matrix-vector multiplication to generate a vector of characteristics, the hollow matrix being generated from a Gabor function which is a sine wave multiplied by a Gaussian function and the Gabor function being a function of at least one variable indicating a position in the normalized matrix and of a set of parameters comprising a parameter relating to the direction of the sine wave, a parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave,
    d) create with the vector of characteristics a probability density for a predetermined list of models,
    e) select the model with the highest probability density as the best model, and
    f) classify the best model as the shape of the input image, in which there are at least two centers of the Gabor function, and in which the wavelength takes at least two values, with a first value of wavelength less than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first value of wavelength is less than a second value of wavelength and greater than or substantially equal to half of the second wavelength value
    BE2015 / 5241 in which the models are characterized by a covariance matrix and by a vector of means, in which the probability density is calculated by the formula p (r) = (2 π. ·· | Ξ F I 2 1 where the symbol r represents the vector of characteristics, the symbol Σ represents the covariance matrix, the symbol μ represents the vector of means and k is equal to the number of elements of the vector of characteristics, in which the set of parameters of the Gabor function includes the standard deviation of the Gaussian function which takes values less than the distance between two adjacent centers of the Gabor function and greater than half the distance between two adjacent centers of the Gabor function.
    [2]
    2. Method according to claim 1, in which all the non-diagonal elements are zero.
    [3]
    3. The method of claim 1, wherein the parameter relating to the direction of the sine wave is an angle.
    [4]
    4. Method according to claim 1, with at least two parameters relating to a center of the Gabor function, in which the parameters relating to a center of the Gabor function are such that the centers of the Gabor function are regularly spaced.
    BE2015 / 5241
    [5]
    5. The method of claim 1, wherein the feature vector is approximated.
    [6]
    6. Computer program product comprising a non-transient computer-readable medium in which a control logic is stored to cause a computer device to identify a shape in an input image, the control logic comprising:
    a) first means of computer-readable program code for normalizing the input image into a standardized matrix representing a normalized image,
    b) second means of computer readable program code for generating an image vector from the standardized matrix,
    c) third computer readable program code means for multiplying the image vector with a sparse matrix using matrix-vector multiplication to generate a feature vector, the sparse matrix (303) being generated from a Gabor function which is a sine wave multiplied by a Gaussian function and the Gabor function being a function of at least one variable indicating a position in the normalized matrix and of a set of parameters including a parameter relating to the direction of the sine wave, a parameter relating to a center of the Gabor function and a parameter relating to a wavelength of the sine wave,
    d) fourth means of computer-readable program code for creating with the feature vector a probability density for a predetermined list of models,
    BE2015 / 5241
    e) fifth means of computer readable program code for selecting the model with the highest probability density as the best model, and
    f) sixth means of computer readable program code for classifying the best model as the shape of the input image, in which there are at least two centers of the Gabor function, and in which the length of wave takes at least two values, with a first wavelength value less than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first wavelength value is less than a second length value wavelength and greater than or substantially equal to half the second wavelength value in which the models are characterized by a covariance matrix and by a vector of means, and in which the probability density is calculated by the formula p (r) =
    Γ (Γ-μ) { Σ (r-μ) · where the symbol r represents the characteristic vector, the symbol Σ represents the covariance matrix, the symbol μ represents the vector of means and k is equal to the number of elements the characteristic vector, in which the set of parameters of the Gabor function comprises the standard deviation of the Gaussian function which takes values less than the distance between two adjacent centers of the Gabor function and greater than half the distance between two adjacent centers of the Gabor function.
    类似技术:
    公开号 | 公开日 | 专利标题
    BE1025504B1|2019-03-27|Pattern recognition system
    BE1025502B1|2019-03-27|SYSTEM AND METHOD FOR FORM RECOGNITION USING GABOR FUNCTIONS
    US20190130232A1|2019-05-02|Font identification from imagery
    BE1024194B1|2017-12-12|Method for identifying a character in a digital image
    BE1022562A9|2017-02-07|Optical character recognition method
    BE1025503B1|2019-03-27|LINE SEGMENTATION METHOD
    FR2974433A1|2012-10-26|EVALUATION OF IMAGE QUALITY
    FR2968426A1|2012-06-08|LARGE SCALE ASYMMETRIC COMPARISON CALCULATION FOR BINARY INTEGRATIONS
    Yang et al.2016|Recapture image forensics based on Laplacian convolutional neural networks
    FR2884007A1|2006-10-06|FACIAL IDENTIFICATION METHOD FROM FACE IMAGES, CORRESPONDING COMPUTER DEVICE AND PROGRAM
    FR3016066A1|2015-07-03|WEIGHTING SYSTEM AND METHOD FOR COLLECTING IMAGE DESCRIPTORS
    EP3570212A1|2019-11-20|Character recognition method
    CA3043090C|2021-07-27|Character recognition process
    EP2839410B1|2017-12-13|Method for recognizing a visual context of an image and corresponding device
    WO2013045593A1|2013-04-04|Improved method of checking the appearance of the surface of a tyre
    CN111274858A|2020-06-12|Business license identification method in network transaction supervision
    Phillips et al.2019|Class Embodiment Autoencoder | for classifying the botanical origins of honey
    FR3018354A1|2015-09-11|METHOD FOR ANALYZING SEDIMENT SAMPLES WITH AUTOMATIC RECOGNITION OF NANNOFOSSILS
    FR3069087B1|2019-08-23|METHOD OF PROCESSING AN IMAGE
    Jiang et al.2014|Log-domain polynomial filters for illumination-robust face recognition
    BE1020588A5|2014-01-07|METHOD OF RECOGNIZING FORMS, COMPUTER PROGRAM PRODUCT, AND MOBILE TERMINAL.
    Rodin et al.2021|Document Image Quality Assessment via Explicit Blur and Text Size Estimation
    BE1021013B1|2014-12-11|METHOD AND SYSTEM FOR IMPROVING THE QUALITY OF COLOR IMAGES
    Pan et al.2021|A Two-Stage Network for Image Deblurring
    EP2856391B1|2019-03-27|Method for detecting drop shadows on an initial image
    同族专利:
    公开号 | 公开日
    CN106462773B|2019-07-12|
    KR20160148563A|2016-12-26|
    WO2015158778A1|2015-10-22|
    US9058517B1|2015-06-16|
    CN106462773A|2017-02-22|
    KR102268174B1|2021-06-24|
    JP2017515221A|2017-06-08|
    JP6609267B2|2019-11-20|
    BE1025502A1|2019-03-20|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US20040017944A1|2002-05-24|2004-01-29|Xiaoging Ding|Method for character recognition based on gabor filters|
    JP4159794B2|2001-05-02|2008-10-01|本田技研工業株式会社|Image processing apparatus and method|
    JP2005044330A|2003-07-24|2005-02-17|Univ Of California San Diego|Weak hypothesis generation device and method, learning device and method, detection device and method, expression learning device and method, expression recognition device and method, and robot device|
    SG121783A1|2003-07-29|2006-05-26|Sony Corp|Techniques and systems for embedding and detectingwatermarks in digital data|
    US8509538B2|2007-12-20|2013-08-13|Intel Corporation|Method and apparatus for obtaining and processing Gabor image features|
    US8391613B2|2009-06-30|2013-03-05|Oracle America, Inc.|Statistical online character recognition|
    CN101866421B|2010-01-08|2013-05-01|苏州市职业大学|Method for extracting characteristic of natural image based on dispersion-constrained non-negative sparse coding|
    CN102915436B|2012-10-25|2015-04-15|北京邮电大学|Sparse representation face recognition method based on intra-class variation dictionary and training image|CN107133622A|2016-02-29|2017-09-05|阿里巴巴集团控股有限公司|The dividing method and device of a kind of word|
    JP6545740B2|2017-03-08|2019-07-17|株式会社東芝|Generating device, program, recognition system and generating method|
    CN108304885A|2018-02-28|2018-07-20|宜宾学院|A kind of Gabor wavelet CNN image classification methods|
    CN108629297A|2018-04-19|2018-10-09|北京理工大学|A kind of remote sensing images cloud detection method of optic based on spatial domain natural scene statistics|
    法律状态:
    2019-05-08| FG| Patent granted|Effective date: 20190327 |
    优先权:
    申请号 | 申请日 | 专利标题
    US14254039|2014-04-16|
    US14/254,039|US9058517B1|2014-04-16|2014-04-16|Pattern recognition system and method using Gabor functions|
    [返回顶部]