WO2005111936A1 - パラメタ推定方法、パラメタ推定装置および照合方法 - Google Patents
パラメタ推定方法、パラメタ推定装置および照合方法 Download PDFInfo
- Publication number
- WO2005111936A1 WO2005111936A1 PCT/JP2005/008006 JP2005008006W WO2005111936A1 WO 2005111936 A1 WO2005111936 A1 WO 2005111936A1 JP 2005008006 W JP2005008006 W JP 2005008006W WO 2005111936 A1 WO2005111936 A1 WO 2005111936A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- learning
- parameter
- face image
- face
- input data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
Definitions
- the present invention relates to a parameter estimating method for estimating a specific parameter with respect to input data force, a parameter estimating apparatus, and a matching method using the same.
- a process of estimating a specific parameter from an input image is a general process in pattern information processing. For example, processing to extract the position of the eyes and nose from the human face image and processing to extract the position of the vehicle image license plate are applicable.
- the parameter estimation method, parameter estimation device, and collation method of the present invention have been made in view of these problems, and have a short processing time, a short processing time, and an accurate parameter for input data at a low processing cost. It is intended to be estimated.
- the parameter estimation method of the present invention expresses an arithmetic method for estimating the maximum posterior probability for input data by using an inner product related to the input data, replacing the inner product with a kernel function, and The feature is to estimate the parameters using the calculation results.
- FIG. 1 is a block diagram of a parameter estimation device according to Embodiment 1 of the present invention.
- FIG. 2 is a flowchart of a parameter estimating apparatus according to Embodiment 1 of the present invention.
- FIG. 3 is a block diagram of a face image matching device according to Embodiment 2 of the present invention.
- FIG. 4 is a flowchart of a face image collating apparatus according to Embodiment 2 of the present invention.
- FIG. 5 is a diagram showing an example of feature points of a face input by a user.
- FIG. 6 is a diagram showing some examples of eye positions extracted in Embodiment 2 of the present invention.
- FIG. 7A is a diagram showing the sensitivity of eye position estimation when the imaging distance of the face image matching device according to Embodiment 2 of the present invention is changed.
- FIG. 7B is a diagram showing sensitivity of eye position estimation when the imaging angle of the face image matching device according to Embodiment 2 of the present invention is changed.
- FIG. 8A is a diagram showing the sensitivity of eye position estimation to the movement of the face position within the screen of a face image photographed with a frontal force.
- FIG. 8B is a diagram showing the sensitivity of eye position estimation with respect to the movement of the face position within the screen of a face image photographed diagonally.
- FIG. 9 is a diagram showing the number of iterations and the sensitivity of eye position estimation.
- FIG. 10 is a block diagram of a face image matching device according to Embodiment 3 of the present invention.
- FIG. 11 is a flowchart of a face image matching device according to Embodiments 3 and 4 of the present invention.
- an operation method for estimating the maximum posterior probability for input data is represented by an inner product related to the input data, the inner product is replaced with a kernel function, and the parameter is estimated using a calculation result of the kernel function. It is characterized by doing. With this method, parameters for input data can be accurately estimated with a short processing time or a small processing cost.
- the parameter estimation method of the present invention includes a step of learning a correlation between a plurality of learning input data whose parameters to be estimated are known and a parameter corresponding to each of the learning input data. And estimating the parameters of the input data for estimation for which the power parameter is unknown using the learned correlation.
- Estimating the parameters of the input data for estimation According to this method, it is possible to accurately estimate parameters for input data with a short processing time or a small processing cost using a computer system.
- the parameter estimating apparatus of the present invention includes: a learning unit that learns a correlation between a plurality of learning input data whose parameters to be estimated are known and a parameter corresponding to each of the learning input data; Estimating means for estimating the parameters of the estimation input data whose parameters to be unknown are learned using the learned correlation, wherein the learning means uses a plurality of learning input data whose parameters to be estimated are known.
- a learning kernel matrix calculating means for calculating a learning kernel matrix; and an eigenvalue calculating means for calculating an eigenvalue and an eigenvector of an inverse matrix of the learning kernel matrix, wherein the estimating means includes an estimating input for which the parameter to be estimated is unknown.
- Estimation kernel calculating means for calculating an estimation kernel using data and input data for learning, a learning kernel matrix, and eigenvalues of an inverse matrix of the learning kernel matrix Characterized in that e Bei and parameter calculation means for calculating the parameters of the estimated input data using its own vector.
- the input data for estimation and the input data for learning may be an image including a face
- the parameter to be estimated may be coordinates of a feature point of the face.
- the matching method of the present invention is a matching method using the parameter estimation device of the present invention and a face image database, wherein the coordinates of the feature points of the face with respect to the face image to be matched using the parameter estimation device. After estimating the face area, the face area is cut out based on the coordinates of the feature points of the face, and the face image registered in the face image database is compared with the cut out face area. This According to the method, it is possible to cut out a face area used for face matching from a face image with a short processing time and a small processing time, and a small processing cost.
- the matching method of the present invention is a matching method using the parameter estimating device of the present invention and a face image database, and estimates the coordinates of the feature points of the face by using the parameter estimating device to obtain the feature of the face.
- the face image normalization operation for normalizing the image area based on the coordinates of the points is repeated a plurality of times, the face area is cut out based on the coordinates of the feature points of the face, and the face image registered in the face image database and the cut out face area are extracted. Collated with the selected face area.
- the input data for estimation and the input data for learning may be an image including a face
- the parameter to be estimated may be height information of the face image.
- a matching method of the present invention is a matching method using the parameter estimating device of the present invention and a face image database, and estimates height information for a face image to be matched using the parameter estimating device. Then, based on the height information of the face image, the two-dimensional image viewed from the same angle as the face image registered in the face image database is synthesized and collated. According to this method, even if the face image faces in an arbitrary direction, it can be compared with the face image database registered as the front or side face image.
- a matching method is a matching method using the parameter estimating device according to claim 7 and a face image database, and is registered in a face image database using the parameter estimating device. After estimating the height information for the face image that is present, a two-dimensional image viewed from an angle other than the registered face image is synthesized and additionally registered in the face image database. According to this method, the input face image is registered in the face image database! Since the face image can be directly collated, high-speed collation can be performed.
- FIG. 1 is a block diagram of a parameter estimating apparatus according to Embodiment 1 of the present invention, which is realized by a computer system.
- the parameter estimating apparatus 100 includes a data input unit 10 for exchanging data with an external device, a CPU 20 for performing data processing, and a program storage unit. And work memory 30, secondary storage (hard disk, magneto-optical disk, etc.) 40 for storing large-scale data such as input data for parameter estimation and correlation data, display 50 as a system console, A mouse 60 is provided as a machine interface, and the above blocks are connected to a system bus 90.
- the CPU 20 functions as a learning kernel matrix calculating unit 22 and an eigenvalue calculating unit 23 of a learning unit by executing a corresponding program, and as an estimated kernel calculating unit 26 and a parameter calculating unit 27 of an estimating unit. work.
- an area for storing various data described later is secured.
- the parameter estimating apparatus 100 is based on the fact that there is a correlation between input data to be estimated and parameters to be estimated. In other words, for many input data for learning, the parameters of which are to be estimated, the correlation between the input data and the parameters is learned at a glance, and any correlation between the input data and the parameters can be made using this correlation.
- the parameter for is estimated.
- the operation of the step of learning the correlation between input data and parameters using a learning sample (hereinafter, referred to as “off-line processing”) will be described.
- the input data is n-dimensional input vector I
- the parameters for input vector I are m-dimensional parameter vectors Q
- the total number of training samples is N.
- FIG. 2 is a flowchart of the parameter estimating apparatus 100 according to Embodiment 1 of the present invention. The following flow is executed by executing the program stored in the CPU 20 and the memory 30.
- the CPU 20 calculates (Equation 2) as follows. Then, a deviation parameter vector from the average parameter vector Q a is obtained, transferred to the secondary storage device 40, and stored (S12).
- the eigenvalue ⁇ and the eigenvector a k are transferred to and stored in the secondary storage device 40 (S15). here
- M indicates the number of independent eigenvectors, and the maximum is equal to the number N of training samples.
- K e represents a deviation learning kernel matrix.
- the data necessary to represent the correlation between the input vector I and the parameter vector Q is further processed in order to simplify the calculation in the estimating step after the uniform force.
- the deviation parameter obtained above Using the tuttle Q e , the deviation learning kernel matrix K e , the eigenvalues and the eigenvectors o; k , (i ij k
- Equation 6 M m-dimensional constant vectors ⁇ are obtained based on Equation 6).
- a secondary storage device 4 0 has an average parameter vector Q a determined in the processing described above, constant vector Omega, constant vector gamma, eigenvectors a k is stored.
- Q a constant vector Omega
- eigenvectors a k is stored.
- online processing a step of actually estimating parameters for input data for which parameters are to be estimated.
- the following flow is executed by the CPU 20 executing the program stored in the memory 30.
- input data for which parameters are to be estimated are input to the data input unit 10 as an input vector I, and stored in the secondary storage device 40 (S 20).
- the obtained estimated kernel K is transferred to the secondary storage device 40 and stored (S21).
- the estimated kernel K obtained in 21 is read from the secondary storage device 40, and a parameter vector Q indicating a parameter to be estimated is calculated using (Equation 9). Then, the estimated parameter parameter Q is stored in the secondary storage device 40 (S22).
- the CPU 20 functions as estimating means.
- the optimal value of the scale ⁇ differs depending on the input vector to be estimated and the parameter to be estimated, which did not mention the scale ⁇ of the Gaussian kernel. Therefore, it is desirable to determine the value while conducting a metametric estimation experiment.
- the present inventors have been studying a MAP (maximum posteriori) estimation method as a method for estimating parameters from image signals. If the simultaneous distribution of the input image I and the parameter Q to be estimated is a random vector following a Gaussian distribution, the optimal estimate of the parameter Q can be obtained using (Equation 10).
- Equation 10 is similar to the equation used for so-called multiple regression analysis. However, if the relationship between the input image I and the parameter Q to be estimated becomes more complicated, and their joint distribution cannot be represented by a Gaussian distribution, then it is difficult to estimate with this simple mathematical formula. It became clear experimentally.
- Kernel trick a new concept which has been introduced into an image recognition technique called a support vector machine and has obtained good results. I paid attention. This is a method of performing non-linear transformation of an input vector and performing linear discrimination in that space.
- the present inventors have studied the introduction of “kernel trick” into the MAP estimation method, and as a result, have established a new parameter estimation method called KMAP. The details are described below.
- the input vector I is nonlinearly transformed using the nonlinear function ⁇ .
- Equation 12 if the non-linearly converted statistic is to be calculated as it is, an enormous calculation in a higher-dimensional space is required. If the operations on the transformed input vector ⁇ can be grouped into the inner product form ⁇ - ⁇ ⁇ ⁇ ⁇ , this can be replaced with the kernel ⁇ ( ⁇ , ⁇ ), greatly reducing the computational complexity. Can be.
- ⁇ [ ⁇ ]
- A diag [l] each sigma _1 eigenvectors, eigenvalues
- the eigenvalue is the deviation learning kernel matrix K e as shown in (Equation 14).
- the expansion coefficient o; k at this time is the k-th eigenvalue of the deviation learning kernel matrix K e
- the average parameter vector Q a , the constant vector ⁇ , the constant vector ⁇ , and the eigenvector a k are obtained in advance using N learning samples, and the input k
- the parameter Q can be estimated using (Equation 6) to (Equation 9).
- a polynomial kernel As the kernel, a polynomial kernel, a sigmoid kernel, a Gaussian kernel, or the like can be used. However, what kind of kernel to use depends on the input vector to be estimated and the parameter to be estimated. It is desirable to determine the kernel while conducting parameter estimation experiments.
- FIG. 3 is a block diagram of a face image matching device according to Embodiment 2 of the present invention.
- the parameter estimating apparatus 100 is configured by a computer system as in the first embodiment, and the blocks are denoted by the same reference numerals as in the first embodiment, and description thereof is omitted.
- a video camera 110 for capturing a face image of a person is connected to the face image matching device 200.
- a face image database 140 in which a face image of a previously registered person is also connected.
- Face image matching apparatus 200 first finds the coordinates of feature points such as eyes, nose, eyebrows, and mouth from the input face image using parameter estimation apparatus 100 in the first embodiment. Next, face image matching apparatus 200 cuts out a face area used for face matching based on the coordinates of the feature points. Specifically, for example, a face area is defined as a square area in which the length of one side is twice as long as the distance between the eyes with the coordinates of the nose as the center, and the upper and lower sides are parallel to a straight line connecting the eyes. Then, the cut out face area is compared with the face image registered in face image database 140.
- a matching method for example, a method such as an eigenface method using principal component analysis, which is a statistical method, can be used.
- the operation of the parameter estimating apparatus 100 that finds the coordinates of the feature points such as eyes, nose, eyebrows, and mouth from the input face image will be described in detail.
- the total number of learning face images is N
- the input vector is an n-dimensional vector in which the values of each pixel of the i-th learning face image are arranged in raster scan order
- the position coordinates of each feature point are m-dimensional parameter vectors Q .
- FIG. 4 is a flowchart of the parameter estimating apparatus 100 used in the face image matching apparatus 200 according to Embodiment 2 of the present invention.
- the camera 110 captures learning face images for N persons.
- the learning face images that is, the learning input vector I.
- FIG. 5 is a diagram showing an example of the facial feature points input by the user.
- the X and Y coordinates of each of the right eyebrow, right eye, left eyebrow, left eye, nose and mouth are input by the user as feature point coordinates.
- the CPU 20 sequentially arranges and connects the coordinate values of each of the feature points input to each of the learning face images to form a learning parameter vector Qi, and stores it in the secondary storage device 40 (S31).
- the CPU 20 calculates the average parameter vector Q a and the constant beta using Equations (1) to (7).
- ⁇ , a constant vector ⁇ , and an eigenvector a k and store them in the secondary storage device 40 (S32 k
- the above is the processing performed by the parameter estimation device 100 offline.
- the data input unit 10 inputs a face image to be collated, arranges the values of each pixel of the face image in raster scan order, converts it into an input vector I, and transfers it to the secondary storage device 40 (S40). .
- the parameter vector Q indicating the parameter to be estimated is calculated using (Equation 9).
- the estimated parameter vector Q is stored in the secondary storage device 40 (S42).
- the CPU 20 decomposes the parameter vector Q into the coordinate data of the feature quantity and displays it on the display together with the input face image (S43). This is the operation to find the coordinates of feature points such as eyes, nose, eyebrows, and mouth from the input face image.
- a face area used for face authentication is cut out based on the coordinates of the feature points, and the cut out face area is compared with a face image registered in the face image database 140.
- FIG. 6 is a diagram showing some examples of eye positions extracted in the second embodiment of the present invention.
- the positions of the extracted eyes are indicated by X marks on the input image.
- FIG. 7 shows the sensitivity of eye position estimation of the face image matching device according to Embodiment 2 of the present invention.
- 7A is a diagram when the photographing distance is changed
- FIG. 7B is a diagram when the photographing angle is changed.
- the horizontal axis shows the estimated eye position error in pixel units
- the vertical axis shows the cumulative extraction rate (correction of the estimated eye position coordinates when the error shown on the horizontal axis is allowed). Therefore, the higher the cumulative extraction rate in the range where the error is small, the higher the sensitivity of position estimation.
- KMAP cumulative extraction rate
- FIG. 8 is a diagram showing the sensitivity of eye position estimation with respect to the movement of the face position within the screen.
- FIG. 8A is for a face image taken from the front
- FIG. 8B is a face taken from an oblique direction. It is for images.
- the moving image was created artificially using the rotation target Gaussian distribution.
- the variance of the movement amount was set to 0, 10, 20, and 30 pixels
- the variance of the rotation angle was 45 degrees
- the average power of the reduction ratio was ⁇ times
- the variance was 0.5 times.
- FIGS. 8A and 8B show the sensitivity of eye position estimation in each method with respect to the variance of each movement amount.
- the method of repeatedly using the KMAP method is as follows. First, the eye position is estimated using the KMAP method, and then the image region is normalized using the estimated eye position. Then, the eye position is estimated again by the KMAP method using the normalized face image. Such a method of repeating the KM AP method n times will be referred to as KMAP (n).
- KMAP (n) KMAP, KMAP (2), KMAP (3), and MLG for a face image having a dispersive power S of 30 pixels.
- the speed of eye position estimation by KMAP was 0.8 seconds per image. According to the MLG method, it was 6 seconds per image (when using a processor equivalent to Pentium IV (registered trademark)). This Thus, according to the KMAP method, not only the estimation accuracy was improved, but also the calculation time was significantly reduced. As described above, it has become possible to accurately determine the feature points of input data with a short processing time and therefore with a small processing cost.
- a face image matching device that estimates height information of a face image using parameter estimation device 100 in the first embodiment will be described.
- FIG. 10 is a block diagram of a face image matching device according to Embodiment 3 of the present invention.
- the face image matching device 300 is configured by a computer system as in the second embodiment.
- the difference from the block diagram of the face image matching device 200 is that two cameras 110 and 115 are provided to capture face images from two directions.
- the face image matching device 300 first estimates the height information of the input face image using the parameter estimating device 100. Next, based on the estimated three-dimensional face image information, a two-dimensional face image viewed from the same angle as the face image registered in the face image database 140 is synthesized and registered. This is for collating with a face image.
- a method of synthesizing a two-dimensional face image from three-dimensional face image information for example, a known method such as rendering in CG (computer graphics) can be used.
- CG computer graphics
- a technique such as a unique face method using principal component analysis, which is a statistical technique, can be applied.
- the operation of estimating the height information of the input face image using the face image matching device 300 will be described.
- the operation of the offline processing for learning the correlation between the input image and the height information of the input image using the learning face image will be described.
- As a method of learning height information it is possible to learn height information independently for each pixel of the input image.However, the input image is divided into a plurality of regions, and the average of each region is calculated. You may learn height information. In this case, the dimension of the parameter vector to be estimated can be reduced, which is more practical.
- the total number of learning face images is N.
- FIG. 11 is a flowchart of the parameter estimating apparatus 100 used in the face image matching apparatus 300 according to Embodiment 3 of the present invention.
- two cameras 110 and 115 that capture faces from different directions capture learning face images for N persons.
- the data input unit 10 transfers and stores these face images to the secondary storage device 40.
- the learning face image captured by one of the cameras 110 is used as a learning input vector I by arranging the values of each pixel in raster scan order (S50).
- the CPU 20 creates three-dimensional information of the face image based on the two learning face images stored in the secondary storage device 40. This is performed using a known method such as rendering using CG.
- the height information of each pixel or each region is sequentially arranged and connected to form one vector, which is referred to as a learning parameter vector Q.
- the learning parameter vector Q is stored in the secondary storage device 40 (S51) 0
- the CPU 20 obtains the average parameter vector Q a , the constant vector ⁇ , the constant vector ⁇ , and the eigenvector a k using (Equation 1) to (Equation 7) and stores them in the secondary storage device 40.
- Equation 1 the average parameter vector Q a , the constant vector ⁇ , the constant vector ⁇ , and the eigenvector a k using (Equation 1) to (Equation 7) and stores them in the secondary storage device 40.
- the above is the processing performed by the parameter estimating apparatus 100 offline.
- the data input unit 10 inputs a face image to be collated, arranges the values of each pixel of the face image in raster scan order, converts the values into an input vector I, and transfers the input vector I to the secondary storage device 40 (S60).
- the average parameter vector Q a , the constant vector ⁇ , the constant vector ⁇ , the eigenvector a k , and the estimated kernel K obtained in step S61, which are obtained by the offline processing, are read from the secondary storage device ki 40, and (Equation 9) ) Is used to calculate the parameter vector Q indicating the parameter to be estimated.
- the estimated parameter vector Q is stored in the secondary storage device 40 (S62).
- the CPU 20 converts the parameter vector into height information of the face image (S63). Up to this is the operation for estimating the height information of the input face image.
- the estimated height information is added to the face image to be collated in this way to obtain three-dimensional face information. Thereafter, as described above, based on the three-dimensional face image information, a two-dimensional face image viewed from the same angle as the face image registered in the face image database 140 is synthesized, and registered with this. ! / Pull the face area used for face authentication based on the coordinates of the feature points that are to be compared with the face image, and match the cut face area with the face image registered in the face image database 140. Collate.
- a front face image or a face image oriented in another direction is synthesized using a two-dimensional face image oriented in an arbitrary direction. can do. Therefore, even if the face image faces in an arbitrary direction, it is possible to collate with the face image database registered as the face image facing forward or sideways.
- Embodiment 3 by estimating the height information of the face image to be compared, the two-dimensional face viewed from the same angle as the face image registered in the face image database is estimated.
- the images were synthesized and collated.
- the height information is estimated for a face image facing the front that has already been registered in the face image database, and face images in several directions are newly registered as a database and collated with them. If a face image to be collated is to be collated, a face image collation device may be configured in the form of ⁇ ⁇ .
- Embodiment 4 will be described below in detail.
- the block diagram of the face image matching device according to the fourth embodiment of the present invention is configured by a computer system as in the third embodiment, and the block diagram is the same as that in the third embodiment, and therefore the description is omitted.
- the operation of the face image matching device according to Embodiment 4 of the present invention will be described below with reference to the flowchart shown in FIG.
- a learning face image for two cameras 110 and a camera 115 that captures faces with different directional forces is taken.
- the data input unit 10 transfers and stores these face images to the secondary storage device 40.
- the learning face image captured by one of the cameras 110 is used as a learning input vector I by arranging the values of each pixel in raster scan order (S50).
- the CPU 20 creates three-dimensional information of the face image based on the two learning face images stored in the secondary storage device 40.
- the height information of each pixel or each region is arranged in order and connected to form one vector, which is referred to as a learning parameter vector Q.
- the learning parameter vector Q is stored in the secondary storage device 40 (S51).
- the CPU 20 calculates (Equation 1) Using Equation 7, the average parameter vector Q a , the constant vector ⁇ , the constant vector ⁇ ,
- the k tuttle a k is obtained and stored in the secondary storage device 40 (S52).
- the data input unit 10 arranges the values of the respective pixels of the front-facing face image already registered in the face image database 140 in the raster scan order, converts them into an input vector I, and converts them into a secondary storage device. Transfer to 40 (S60).
- the parameter is read from the secondary storage device 40, the parameter parameter Q indicating the parameter to be estimated is calculated using (Equation 9), and stored in the secondary storage device 40 (S62).
- the CPU 20 converts the parameter vector into height information of the face image.
- the estimated height information is added to the registered face image to obtain three-dimensional face information (S63).
- two-dimensional face images viewed from angles such as right, diagonally right, diagonally left, diagonally left, diagonally upward, and diagonally downward are synthesized.
- the composite image of is newly registered in the face image database 140.
- the above processing is performed on each of the registered face images that are not powerful when viewed from one direction, and the two-dimensional face images viewed from each direction are synthesized and registered as a database.
- the face image matching device 300 uses the camera 110 to capture a face image to be compared. Then, a face area used for face authentication is cut out based on the coordinates of the feature point to be checked, and the cut out face area is checked against a face image registered in face image database 140.
- the number of face images to be registered in the face image database increases, but the face image and the face image database input in online processing! Since it is possible to directly collate with the face images registered in the source, the collation can be performed at high speed.
- the operation shown in the flowchart is performed by the CPU reading the program.
- a configuration including a dedicated processor for executing learning means, learning kernel matrix calculating means, eigenvalue calculating means, estimating means, estimated kernel calculating means, parameter calculating means, and the like may be adopted.
- a computer program product that is a storage medium including instructions that can be used to program a computer that implements the present invention is included in the scope of the present invention.
- These storage media include disks such as flexible disks, optical disks, CDROMs, and magnetic disks, ROMs, RAMs, EPROMs, EEPROMs, magneto-optical cards, memory cards, DVDs, and the like.
- Embodiments 2 to 4 an example in which the parameter estimation device of the present invention is used for image matching is shown.
- Force data other than an image is input as force input data to be used for a matching device other than image matching.
- be able to For example, by inputting voice data as input data, it can be used for a voice verification device.
- a parameter estimating method it is possible to provide a parameter estimating method, a parameter estimating device, and a matching method that can accurately estimate parameters for input data with a short processing time or a small processing cost.
- the present invention can provide a parameter estimation method, a parameter estimation device, and a collation method that can accurately estimate parameters for input data with a short processing time or a small processing cost. This is effective for a parameter estimation method for estimating a parameter, a parameter estimation device, and a matching method using the same.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Human Computer Interaction (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
- Image Processing (AREA)
Abstract
Description
Claims
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US11/596,431 US7760933B2 (en) | 2004-05-14 | 2005-04-27 | Parameter estimation method, parameter estimation device and collation method |
| EP05737145.2A EP1758059A4 (en) | 2004-05-14 | 2005-04-27 | PARAMETER METHOD, PARAMETER EQUIPMENT AND COLLATION METHOD |
| CN2005800155586A CN1954342B (zh) | 2004-05-14 | 2005-04-27 | 参数推定方法、参数推定装置及对照方法 |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2004-144504 | 2004-05-14 | ||
| JP2004144504A JP4321350B2 (ja) | 2004-05-14 | 2004-05-14 | パラメタ推定方法、パラメタ推定装置および照合方法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2005111936A1 true WO2005111936A1 (ja) | 2005-11-24 |
Family
ID=35394356
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2005/008006 Ceased WO2005111936A1 (ja) | 2004-05-14 | 2005-04-27 | パラメタ推定方法、パラメタ推定装置および照合方法 |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US7760933B2 (ja) |
| EP (1) | EP1758059A4 (ja) |
| JP (1) | JP4321350B2 (ja) |
| CN (1) | CN1954342B (ja) |
| WO (1) | WO2005111936A1 (ja) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102314614A (zh) * | 2011-10-24 | 2012-01-11 | 北京大学 | 一种基于类共享多核学习的图像语义分类方法 |
| EP2441857A2 (en) | 2008-10-23 | 2012-04-18 | P2i Ltd | Vacuum processing apparatus |
Families Citing this family (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8638846B1 (en) * | 2003-06-23 | 2014-01-28 | At&T Intellectual Property Ii, L.P. | Systems and methods for encoding and decoding video streams |
| JP4939968B2 (ja) * | 2007-02-15 | 2012-05-30 | 株式会社日立製作所 | 監視画像処理方法、監視システム及び監視画像処理プログラム |
| JP4837602B2 (ja) * | 2007-03-12 | 2011-12-14 | 富士フイルム株式会社 | 画像トリミング装置および方法並びにプログラム |
| KR20250099420A (ko) | 2007-09-24 | 2025-07-01 | 애플 인크. | 전자 장치 내의 내장형 인증 시스템들 |
| US8600120B2 (en) | 2008-01-03 | 2013-12-03 | Apple Inc. | Personal computing device control using face detection and recognition |
| US8208731B2 (en) * | 2008-04-07 | 2012-06-26 | Microsoft Corporation | Image descriptor quantization |
| JP5418061B2 (ja) * | 2009-08-20 | 2014-02-19 | 沖電気工業株式会社 | 遠隔コミュニケーション装置、遠隔コミュニケーション方法、および遠隔コミュニケーションシステム |
| JP5923723B2 (ja) * | 2011-06-02 | 2016-05-25 | パナソニックIpマネジメント株式会社 | 人物属性推定システム、人物属性推定装置、及び人物属性推定方法 |
| US8638385B2 (en) | 2011-06-05 | 2014-01-28 | Apple Inc. | Device, method, and graphical user interface for accessing an application in a locked device |
| US9002322B2 (en) | 2011-09-29 | 2015-04-07 | Apple Inc. | Authentication with secondary approver |
| EP2850510A2 (en) | 2012-05-18 | 2015-03-25 | Apple Inc. | Device, method, and graphical user interface for manipulating user interfaces based on fingerprint sensor inputs |
| CN103413115B (zh) * | 2013-06-06 | 2016-09-28 | 温州大学 | 一种基于数据流形分布的非线性图像特征选择方法 |
| US9898642B2 (en) | 2013-09-09 | 2018-02-20 | Apple Inc. | Device, method, and graphical user interface for manipulating user interfaces based on fingerprint sensor inputs |
| US20150133798A1 (en) * | 2013-11-13 | 2015-05-14 | The Regents Of The University Of California | Methods for noninvasive intracranial pressure calibration without the need for invasive icp |
| US9324067B2 (en) | 2014-05-29 | 2016-04-26 | Apple Inc. | User interface for payments |
| CN105306097B (zh) * | 2014-06-18 | 2019-07-26 | 中兴通讯股份有限公司 | 一种随机接入信号的检测方法、装置和系统 |
| KR102077260B1 (ko) * | 2014-08-08 | 2020-02-13 | 삼성전자주식회사 | 확룔 모델에 기반한 신뢰도를 이용하여 얼굴을 인식하는 방법 및 장치 |
| US9762791B2 (en) * | 2014-11-07 | 2017-09-12 | Intel Corporation | Production of face images having preferred perspective angles |
| DK179186B1 (en) | 2016-05-19 | 2018-01-15 | Apple Inc | REMOTE AUTHORIZATION TO CONTINUE WITH AN ACTION |
| DK179471B1 (en) | 2016-09-23 | 2018-11-26 | Apple Inc. | IMAGE DATA FOR ENHANCED USER INTERACTIONS |
| KR102185854B1 (ko) | 2017-09-09 | 2020-12-02 | 애플 인크. | 생체측정 인증의 구현 |
| KR102389678B1 (ko) | 2017-09-09 | 2022-04-21 | 애플 인크. | 생체측정 인증의 구현 |
| US10949497B2 (en) * | 2018-01-26 | 2021-03-16 | Crocus Energy | High-speed multi-input tracker based on in-memory operations of time-dependent data |
| US11170085B2 (en) | 2018-06-03 | 2021-11-09 | Apple Inc. | Implementation of biometric authentication |
| US11295223B2 (en) | 2018-06-12 | 2022-04-05 | International Business Machines Corporation | Quantum feature kernel estimation using an alternating two layer quantum circuit |
| US10860096B2 (en) | 2018-09-28 | 2020-12-08 | Apple Inc. | Device control using gaze information |
| US11100349B2 (en) | 2018-09-28 | 2021-08-24 | Apple Inc. | Audio assisted enrollment |
| EP4264460B1 (en) | 2021-01-25 | 2025-12-24 | Apple Inc. | Implementation of biometric authentication |
| WO2022185644A1 (ja) * | 2021-03-03 | 2022-09-09 | 富士フイルム株式会社 | 画像処理装置、画像処理装置の作動方法、画像処理装置の作動プログラム、並びに学習方法 |
| US12210603B2 (en) | 2021-03-04 | 2025-01-28 | Apple Inc. | User interface for enrolling a biometric feature |
| US12216754B2 (en) | 2021-05-10 | 2025-02-04 | Apple Inc. | User interfaces for authenticating to perform secure operations |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000090274A (ja) * | 1998-09-16 | 2000-03-31 | Nippon Telegr & Teleph Corp <Ntt> | パターン識別方法及びパターン識別プログラムを記録した記録媒体 |
| JP2001266145A (ja) * | 2000-03-22 | 2001-09-28 | Matsushita Electric Ind Co Ltd | パラメタ推定方法及びその装置 |
| JP2003173444A (ja) * | 2001-09-25 | 2003-06-20 | Matsushita Electric Ind Co Ltd | 複合照合装置 |
| JP2004030552A (ja) * | 2001-09-25 | 2004-01-29 | Matsushita Electric Ind Co Ltd | パラメタ推定装置及びデータ照合装置 |
-
2004
- 2004-05-14 JP JP2004144504A patent/JP4321350B2/ja not_active Expired - Fee Related
-
2005
- 2005-04-27 CN CN2005800155586A patent/CN1954342B/zh not_active Expired - Fee Related
- 2005-04-27 EP EP05737145.2A patent/EP1758059A4/en not_active Withdrawn
- 2005-04-27 US US11/596,431 patent/US7760933B2/en not_active Expired - Fee Related
- 2005-04-27 WO PCT/JP2005/008006 patent/WO2005111936A1/ja not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000090274A (ja) * | 1998-09-16 | 2000-03-31 | Nippon Telegr & Teleph Corp <Ntt> | パターン識別方法及びパターン識別プログラムを記録した記録媒体 |
| JP2001266145A (ja) * | 2000-03-22 | 2001-09-28 | Matsushita Electric Ind Co Ltd | パラメタ推定方法及びその装置 |
| JP2003173444A (ja) * | 2001-09-25 | 2003-06-20 | Matsushita Electric Ind Co Ltd | 複合照合装置 |
| JP2004030552A (ja) * | 2001-09-25 | 2004-01-29 | Matsushita Electric Ind Co Ltd | パラメタ推定装置及びデータ照合装置 |
Non-Patent Citations (3)
| Title |
|---|
| IZOE S. ET AL.: "Gazo no Fractal-sei ni Motozuita Kernel Shuseibun Bunseki ni yoru Shahei Ryoiki no Gengazo Suitei", THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS GIJUTSU KENKYU HOKOKU, vol. 101, no. 362, 11 October 2001 (2001-10-11), pages 55 - 62, XP008114986 * |
| MAEDA E.: "Tsukai! Support Vector Machine", JOHO SHORI, IPSJ MAGAZINE, 15 July 2001 (2001-07-15), pages 676 - 683, XP008114987 * |
| TSUDA H.: "Support Vector Machine towa Nanika", THE JOURNAL OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS, vol. 83, no. 6, 25 June 2000 (2000-06-25), pages 460 - 466, XP008114989 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2441857A2 (en) | 2008-10-23 | 2012-04-18 | P2i Ltd | Vacuum processing apparatus |
| CN102314614A (zh) * | 2011-10-24 | 2012-01-11 | 北京大学 | 一种基于类共享多核学习的图像语义分类方法 |
| CN102314614B (zh) * | 2011-10-24 | 2013-06-05 | 北京大学 | 一种基于类共享多核学习的图像语义分类方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| US7760933B2 (en) | 2010-07-20 |
| JP4321350B2 (ja) | 2009-08-26 |
| EP1758059A1 (en) | 2007-02-28 |
| US20070230773A1 (en) | 2007-10-04 |
| CN1954342A (zh) | 2007-04-25 |
| EP1758059A4 (en) | 2013-08-21 |
| JP2005327076A (ja) | 2005-11-24 |
| CN1954342B (zh) | 2010-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP4321350B2 (ja) | パラメタ推定方法、パラメタ推定装置および照合方法 | |
| CN100472559C (zh) | 特征点检测装置、特征点检测方法 | |
| Bhagavatula et al. | Faster than real-time facial alignment: A 3d spatial transformer network approach in unconstrained poses | |
| US9053388B2 (en) | Image processing apparatus and method, and computer-readable storage medium | |
| JP5406705B2 (ja) | データ補正装置及び方法 | |
| US10755145B2 (en) | 3D spatial transformer network | |
| US7925060B2 (en) | Authentication system and registration system related to facial feature information | |
| JP4595750B2 (ja) | 画像処理装置および方法、並びにプログラム | |
| EP2353126B1 (en) | Keypoint descriptor generation by complex wavelet analysis | |
| JP4466951B2 (ja) | 立体結合顔形状の位置合わせ | |
| CN109684969B (zh) | 凝视位置估计方法、计算机设备及存储介质 | |
| CN107016319A (zh) | 一种关键点定位方法和装置 | |
| Sung et al. | Pose-Robust Facial Expression Recognition Using View-Based 2D $+ $ 3D AAM | |
| CN109934926B (zh) | 模型数据处理方法、装置、可读存储介质和设备 | |
| JP2021503139A (ja) | 画像処理装置、画像処理方法および画像処理プログラム | |
| CN117372604A (zh) | 一种3d人脸模型生成方法、装置、设备及可读存储介质 | |
| CN112990047A (zh) | 一种结合面部角度信息的多姿态人脸验证方法 | |
| JP3476710B2 (ja) | ユークリッド的な3次元情報の復元方法、および3次元情報復元装置 | |
| WO2015176502A1 (zh) | 一种图像特征的估计方法和设备 | |
| CN112488058A (zh) | 面部跟踪方法、装置、设备和存储介质 | |
| US12573162B2 (en) | Modelling method for making a virtual model of a user's head | |
| JP7623994B2 (ja) | 画像処理装置、画像処理方法、及びプログラム | |
| JP4068888B2 (ja) | アイリス推定装置およびアイリス照合装置 | |
| CN103310219B (zh) | 配准对象形状的精度评估方法和设备、配准的方法和设备 | |
| Fan et al. | 3D hand skeleton model estimation from a depth image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS KE KG KM KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SM SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW |
|
| AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LT LU MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
| WWE | Wipo information: entry into national phase |
Ref document number: 2005737145 Country of ref document: EP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 11596431 Country of ref document: US Ref document number: 2007230773 Country of ref document: US Ref document number: 200580015558.6 Country of ref document: CN |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWW | Wipo information: withdrawn in national office |
Country of ref document: DE |
|
| WWP | Wipo information: published in national office |
Ref document number: 2005737145 Country of ref document: EP |
|
| WWP | Wipo information: published in national office |
Ref document number: 11596431 Country of ref document: US |



