WO2015029982A1 - 画像処理装置、画像処理方法、及びプログラム - Google Patents
画像処理装置、画像処理方法、及びプログラム Download PDFInfo
- Publication number
- WO2015029982A1 WO2015029982A1 PCT/JP2014/072258 JP2014072258W WO2015029982A1 WO 2015029982 A1 WO2015029982 A1 WO 2015029982A1 JP 2014072258 W JP2014072258 W JP 2014072258W WO 2015029982 A1 WO2015029982 A1 WO 2015029982A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- face
- image
- geometric deformation
- parameters
- parameter
- Prior art date
Links
- 238000012545 processing Methods 0.000 title claims abstract description 92
- 238000003672 processing method Methods 0.000 title claims description 13
- 238000006243 chemical reaction Methods 0.000 claims abstract description 49
- 239000007787 solid Substances 0.000 claims description 74
- 239000011159 matrix material Substances 0.000 claims description 42
- 230000009466 transformation Effects 0.000 claims description 38
- 238000004364 calculation method Methods 0.000 claims description 30
- 238000012937 correction Methods 0.000 claims description 20
- 230000036544 posture Effects 0.000 description 94
- 238000000034 method Methods 0.000 description 37
- 230000008569 process Effects 0.000 description 22
- 238000001514 detection method Methods 0.000 description 21
- 238000010586 diagram Methods 0.000 description 8
- 239000013598 vector Substances 0.000 description 8
- 239000000284 extract Substances 0.000 description 5
- 238000013500 data storage Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000005477 standard model Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06V40/172—Classification, e.g. identification
Definitions
- the present invention relates to an image processing apparatus, an image processing method, and a program.
- a method of generating and collating a face image different from the input face image by using a three-dimensional model of the face is disclosed.
- Patent Document 1 acquires a two-dimensional image of a person as a subject and three-dimensional face shape information, a standard model that is a three-dimensional model of a general face prepared in advance, and the acquired three-dimensional Two-dimensional using two-dimensional features obtained by correcting a two-dimensional image texture based on posture / shape change information between face shape information and three-dimensional feature points obtained from three-dimensional face shape information
- a method for matching a person in an image is disclosed.
- Patent Document 2 after estimating a face posture using stable feature points irrespective of a person, a difference is noticeable for each person using a standard face solid shape model.
- a method is disclosed in which the feature points are converted into specific postures and collated by comparing the feature amounts at the positions of the other feature points.
- Patent Document 3 based on feature points arranged on a standard three-dimensional model of a face, face feature points in an input image, and face orientation information in the input image, A technique is disclosed in which a partial region image obtained by deforming a partial region around a feature point so as to have the same face orientation as that of a reference image is generated, and the partial region image is compared with the reference image.
- Patent Document 1 it is necessary to acquire three-dimensional shape information of a person to be collated. Since special equipment such as a range finder is required to acquire the three-dimensional shape information of the object, the use environment may be limited. Furthermore, it is necessary not only to acquire 3D shape information, but also to extract each feature quantity from both the 2D image texture and the 3D shape information, and the calculation cost is high.
- Patent Document 2 when estimating the posture based on the correspondence between the feature point coordinates of the face in the image and the feature point coordinates on the standard face three-dimensional model, both eyes, Only 4 points of nose and mouth are used. Therefore, there is a high possibility that a wrong posture is estimated when there is an erroneous detection of feature points or an outlier.
- Patent Document 3 when the posture is converted using the face orientation information acquired in advance by some means, the posture is converted only for a partial region around the feature point. It is greatly affected by detection. In addition, it is considered that the method is not suitable when the global feature of the entire face is used for matching.
- the present invention has been made in view of the above problems, and an object of the present invention is to provide an image processing apparatus, an image processing method, and a program that can generate a normalized image with high accuracy without special equipment. It is in.
- Posture estimation means for estimating posture information including a yaw angle and a pitch angle of the person's face from an input image including the person's face; The position of a plurality of feature points in a face area image that is an area including the face of the person in the input image, the position of the plurality of feature points in a three-dimensional solid model of the face of the person, and the posture information
- An image conversion means for generating a normalized face image in which the orientation of the face is corrected,
- An image processing apparatus is provided.
- Computer Posture estimation means for estimating posture information including a yaw angle and a pitch angle of the human face from an input image including the human face; The position of a plurality of feature points in a face area image that is an area including the face of the person in the input image, the position of the plurality of feature points in a three-dimensional solid model of the face of the person, and the posture information
- An image conversion means for generating a normalized face image in which the orientation of the face is corrected,
- a program for functioning as a server is provided.
- FIG. 3 is a flowchart illustrating a processing flow of the image processing apparatus according to the first embodiment. It is a block diagram which shows the detailed process structure of an image conversion part. It is a block diagram which shows the process structural example of the image processing apparatus in 2nd Embodiment. It is a flowchart which shows the flow of a process of the image processing apparatus in 2nd Embodiment. It is a flowchart which shows the detailed flow of a process of S106. It is a block diagram which shows the detailed process structure of the image conversion part in 3rd Embodiment. It is a flowchart which shows the detailed flow of a process of S106 in 3rd Embodiment.
- FIG. 1 is a block diagram illustrating a configuration example of an image processing apparatus 10 according to the first embodiment.
- the image processing apparatus 10 includes a posture estimation unit 110 and an image conversion unit 120.
- the posture estimation unit 110 estimates the posture information of the human face from the input image including the human face.
- the “posture information” is information indicating the posture of a person's face, and includes at least the yaw angle and pitch angle of the person's face. Posture information can also be said to be information representing the orientation of a person's face.
- the posture estimation unit 110 can estimate posture information of a person's face in the input image using various means. For example, a face discriminator corresponding to each posture having different yaw angles and pitch angles is prepared in a storage unit (not shown), and the posture estimation unit 110 includes a face region image including at least a human face in the input image. And the face information including the yaw angle and pitch angle of the person's face in the input image can be estimated.
- the posture estimation unit 110 may estimate the posture of a person in the input image using a subspace method.
- the posture estimation unit 110 acquires a face area image from the input image and simultaneously acquires a three-dimensional shape of the face, the acquired three-dimensional shape of the face, a storage area of the image processing apparatus 10, and the like.
- the posture of the person's head in the input image may be estimated from the correspondence relationship with the average three-dimensional shape of the front face stored in advance.
- an ICP Intelligent Closest Point
- an ICP Intelligent Closest Point
- details of a method for estimating posture information are disclosed in, for example, the following documents. ⁇ Ando, Kusachi, et al., “Pose Estimation Method of 3D Objects Using Support Vector Regression”, IEICE Transactions ⁇ Yamada, Nakajima, et al.
- posture estimation unit 110 may It is also possible to estimate posture information of a person's face using a known method.
- the image conversion unit 120 is based on the positions of the plurality of feature points in the face area image, the positions of the plurality of feature points in the three-dimensional solid shape model of the person's face, and the posture information acquired by the posture estimation unit 110. Then, a normalized face image in which the face orientation is corrected is generated.
- a normalized face image in which the face orientation is corrected is generated.
- the image conversion unit 120 uses a three-dimensional solid model whose face is facing the front.
- a normalized face image facing front will be described as an example.
- the image conversion unit 120 generates a normalized face image using the face area image and the 3D solid shape model stored in the 3D model storage unit 210 of the storage unit 20.
- the three-dimensional model storage unit 210 stores an average front face three-dimensional solid shape model and positions of a plurality of feature points in the three-dimensional solid shape model.
- This “three-dimensional model of an average front face” can be generated by, for example, averaging a plurality of face image samples facing the front.
- a plurality of “average front face three-dimensional solid model” may be prepared for each attribute such as age or sex.
- the image processing apparatus 10 includes the storage unit 20, but another apparatus located outside the image processing apparatus 10 may include the storage unit 20. In this case, the image processing apparatus 10 acquires a three-dimensional solid model by communicating with the other apparatus by wire or wireless.
- the image conversion unit 120 generates a normalized face image as follows. First, the image conversion unit 120 determines a correspondence relationship between a plurality of feature points on the three-dimensional solid shape model and a plurality of feature points on the face area image. Next, the image conversion unit 120 uses the posture information acquired by the posture estimation unit 110 so that the posture (face orientation) of the three-dimensional model of the face is the same as the posture of the face area image. Rotate the 3D solid shape model. Then, the image conversion unit 120 calculates the coordinate system of the input image from the correspondence between the positions of the plurality of feature points of the three-dimensional solid shape model corrected to the same posture and the positions of the plurality of feature points in the face area image. A geometric deformation parameter for converting the coordinate system of the three-dimensional solid shape model is calculated. Then, the image conversion unit 120 calculates coordinates when the three-dimensional model of the front face is projected onto the coordinate system of the input image using the calculated geometric deformation parameter.
- a front face three-dimensional solid model is composed of a plurality of points, and each point constituting the three-dimensional solid model corresponds to one pixel of a normalized face image.
- the image conversion unit 120 uses the calculated geometric deformation parameter to project (forward projection) each point constituting the three-dimensional solid shape model onto the two-dimensional input image, thereby generating a three-dimensional solid shape model.
- the color information (pixel value) that each point should have can be determined.
- the image conversion unit 120 has pixels that each point of the 3D shape model has pixel values corresponding to positions on the 2D image on which the points constituting the 3D solid shape model are projected. Judge as value.
- the image conversion unit 120 can determine the pixel value of each pixel of the normalized face image based on the correspondence between each point of the three-dimensional solid shape model and the normalized face image.
- the image conversion process performed by the image conversion unit 120 is not a forward conversion process but an inverse conversion process. Specifically, the image conversion unit 120 determines which part on the two-dimensional image each coordinate of the normalized face image corresponds to through the three-dimensional solid model, and determines each pixel of the normalized face image. Pixel values are acquired from a two-dimensional image (inverse conversion process).
- the image conversion unit 120 determines which part of the normalized face image each coordinate of the two-dimensional image corresponds to, and converts the pixel value corresponding to each coordinate of the two-dimensional face image into the normalized face image.
- a more accurate normalized face image can be generated than in the case of using the forward conversion process to be embedded.
- the image conversion unit 120 performs normalization in which the position and size of the face and the orientation of the face are corrected to be constant by performing an inverse conversion process on each point constituting the three-dimensional solid shape model.
- a face image can be generated.
- the coordinate value when each point of the three-dimensional solid shape model is projected on the two-dimensional image based on the geometric deformation parameter is not necessarily an integer value.
- the coordinate value projected on the two-dimensional image is a decimal value, it is desirable to interpolate the pixel value of each pixel of the normalized face image using peripheral pixels of the projected coordinate.
- the image conversion unit 120 interpolates the pixel value of each pixel of the normalized face image using an arbitrary method such as nearest neighbor interpolation (nearest neighbor interpolation) or bilinear interpolation (bilinear interpolation). can do.
- each component of the image processing apparatus 10 shown in the drawing is not a hardware unit configuration but a functional unit block.
- Each component of the image processing apparatus 10 is centered on an arbitrary computer CPU, memory, a program that implements the components shown in the figure loaded in the memory, a storage medium such as a hard disk for storing the program, and a network connection interface. It is realized by any combination of hardware and software. There are various modifications of the implementation method and apparatus.
- FIG. 2 is a flowchart showing a processing flow of the image processing apparatus 10 according to the first embodiment.
- the image processing apparatus 10 extracts a face area image from the input image (S102).
- the extracted face area image only needs to include the face of a person in the input image, and may be the input image itself or a part of the input image.
- the image processing apparatus 10 estimates the posture information of the face of the person included in the face area image (S104).
- the image processing apparatus 10 determines the correspondence between the positions of the plurality of feature points in the extracted face region image and the positions of the plurality of feature points in the three-dimensional model of the face, and the posture estimated in S104. Based on the information, a normalized face image with the face orientation corrected is generated (S106).
- the three-dimensional model of the face is rotated to the same posture (face orientation) as the face of the person in the input image.
- the coordinate axis of the input image and the coordinate axis of the three-dimensional solid model can be converted from each other from the correspondence between the positions of the plurality of feature points in the input image and the positions of the feature points of the rotated three-dimensional solid model.
- a geometric deformation parameter is calculated.
- a normalized face image in which the face orientation is corrected is generated using the calculated geometric deformation parameter.
- the present embodiment when estimating the geometric deformation parameter, it is possible to reduce the influence of feature points whose correct positions cannot be detected due to erroneously detected feature points or occlusions. That is, a more reliable geometric deformation parameter can be estimated. Then, by using the geometric deformation parameter estimated in this way, a normalized face image with higher accuracy can be generated. For example, even for face images that are difficult to normalize because some feature points are hidden by the posture of the person's face in the input image, the hidden feature points can be accurately detected by matching the 3D solid shape model to the posture of the input image. It is possible to estimate well and to generate a normalized face image with higher accuracy.
- the normalized face image generated by the image processing apparatus 10 is converted into a state in which the face position, size, and orientation are constant.
- the position, size, and orientation of the face in the normalized face image are set so as to match the face location, size, and orientation of the correct answer data used in the face authentication matching process, the matching process is performed. Accuracy can be improved.
- the image processing apparatus 10 In the present embodiment, a detailed processing configuration of the image conversion unit 120 will be described.
- the image processing apparatus 10 according to the present embodiment further includes a configuration for collating face images.
- the image processing apparatus 10 generally operates as follows. First, the image processing apparatus 10 calculates a geometric deformation parameter from the correspondence between the position of each feature point in the input face image and the position on the three-dimensional solid shape model. Then, the image processing apparatus 10 optimizes (corrects) the geometric deformation parameter so that the square sum of the reprojection error is minimized.
- the yaw angle and pitch angle estimated from the input image are used as the yaw angle and pitch angle. Since the yaw angle and pitch angle estimated from the input image are more accurate than the yaw angle and pitch angle calculated from the correspondence between the positions of the feature points, the image processing apparatus 10 can perform geometric deformation with higher accuracy.
- the parameter can be estimated.
- FIG. 3 is a block diagram illustrating a detailed processing configuration of the image conversion unit 120.
- the image conversion unit 120 includes a parameter calculation unit 122, a parameter correction unit 124, and a normalized face image generation unit 126.
- the parameter calculation unit 122 calculates the coordinate system of the input image and the three-dimensional solid model. A geometric deformation parameter that can be converted to the coordinate system is calculated.
- the parameter correction unit 124 corrects the geometric deformation parameter calculated by the parameter calculation unit 122 based on the posture information estimated by the posture estimation unit 110.
- the normalized face image generation unit 126 generates a normalized face image with the face orientation corrected based on the geometric deformation parameters corrected by the parameter correction unit 124.
- FIG. 4 is a block diagram illustrating a processing configuration example of the image processing apparatus 10 according to the second embodiment.
- the image processing apparatus 10 further includes a face detection unit 130, a face feature point detection unit 140, and a face identification unit 150.
- the storage unit 20 further includes a collation data storage unit 220.
- the collation data storage unit 220 stores collation data used to collate the person of the normalized face image.
- the “collation data” refers to data in which information indicating a specific person is associated with certain face image data. That is, it can be said that the collation data is face image data that can identify a certain person.
- the face detection unit 130 detects and extracts an area including a face (face area image) from the input image.
- the face detection unit 130 can use any face detection algorithm.
- the face detection unit 130 may use a face detection algorithm or the like that uses Haar features and AdaBoost proposed by Viola et al.
- the face area image detected by the face detection unit 130 is input to the posture estimation unit 110 and the face feature point detection unit 140.
- the face feature point detection unit 140 detects a feature point from the face image area detected by the face detection unit 130.
- the face feature point detection unit 140 detects eyes, nose, mouth, face outline, and the like as feature points.
- the face feature point detection unit 140 can use any algorithm that detects feature points from the face image region.
- the face feature point detection unit 140 can use a feature point detection algorithm using Haar features and AdaBoost, as with the face detection unit 130.
- the face feature point detection unit 140 may detect the feature points of the face using ActiveAShape Model, Active Appearance Model, or the like.
- the face identification unit 150 extracts a feature amount from the normalized face image generated by the image conversion unit 120, and uses the extracted feature amount and the feature amount of the matching data stored in the matching data storage unit 220. The person of the input image is specified by collating. Then, the face identification unit 150 outputs the identified person as a matching result.
- FIG. 5 is a flowchart showing the flow of processing of the image processing apparatus 10 in the second embodiment.
- the process flow from S102 to S106 is as described in the first embodiment.
- the image processing apparatus 10 extracts a face area image including a human face from the input image using an arbitrary face detection algorithm (S102). Then, the image processing apparatus 10 estimates the posture information of the face of the person included in the face area image (S104).
- FIG. 6 is a flowchart showing a detailed flow of the process of S106.
- the image processing apparatus 10 detects the positions (u n , v n ) of n feature points (n is an integer of 1 or more) from the face area image extracted in S102 using an arbitrary feature point detection algorithm. (S1061).
- the image processing apparatus 10 is based on the correspondence between the feature points (u n , v n ) detected in S1061 and the feature points (X n , Y n , Z n ) in the three-dimensional solid shape model.
- Geometric deformation parameters are calculated (S1062). Specifically, the image processing apparatus 10 calculates the geometric deformation parameter as follows.
- the following Expression 1 is an expression showing the correspondence between the coordinates on the input image and the coordinates on the three-dimensional solid shape model.
- a matrix represented by 3 rows and 4 columns is a perspective projection transformation matrix for transforming the coordinate system on the input image and the coordinate system on the three-dimensional solid shape model.
- the image processing apparatus 10 uses the feature points (u n , v n ) detected in the face area image and the feature points (X n , Y n ) in the three-dimensional solid model corresponding to the feature points (u n , v n ). , Z n ) and Equation 1 below, the geometric deformation parameters included in the perspective projection transformation matrix are derived.
- a perspective projection transformation matrix of 3 rows and 4 columns can be obtained by using the linear least square method with respect to Equation 2.
- the perspective projection transformation matrix can be decomposed as shown in Equation 3 below.
- Equation 3 “K” is a matrix representing the internal parameters of the camera, and its degree of freedom is 5. “R” and “T” are matrices representing the external parameters of the camera, respectively, and the degree of freedom is 6. “R” is an external parameter relating to rotation, “ ⁇ ” is a yaw angle, “ ⁇ ” is a pitch angle, and “ ⁇ ” is a roll angle. “T” is an external parameter related to the translation component.
- the geometric deformation parameters included in the perspective projection transformation matrix have a total of 11 degrees of freedom.
- Equation 2 parameters may be calculated by eigenvalue calculation or the like instead of the linear least square method.
- each feature point is projected by performing coordinate transformation in which the origin of the coordinate system of each feature point in the input image and the origin and scale of the coordinate system of each feature point on the three-dimensional solid model are aligned in advance.
- Geometric deformation parameters may be calculated so that the coordinates at that time are approximately appropriate.
- the image processing apparatus 10 corrects the geometric deformation parameter using the posture information (yaw angle, pitch angle) estimated in S104 (S1063). Specifically, the image processing apparatus 10 corrects the geometric deformation parameter as follows.
- the image processing apparatus 10 sets the yaw angle ⁇ and the pitch angle ⁇ as fixed values among the 11 geometric deformation parameters shown in Expression 3, and minimizes the sum of squares of the reprojection error for the remaining nine parameters. Optimize to.
- the yaw angle ⁇ and the pitch angle ⁇ the yaw angle and pitch angle included in the posture information estimated in S104 are used.
- the geometric deformation parameters calculated in S1062 are used as initial values.
- the image processing apparatus 10 substitutes the yaw angle and pitch angle included in the posture information estimated in S104 as initial values in the eleven geometric deformation parameters shown in Expression 3, and is included in the perspective projection transformation matrix.
- the eleven parameters may be optimized so that the sum of squares of the reprojection error is minimized.
- the yaw angle ⁇ and the pitch angle ⁇ are limited to be optimized within a predetermined range based on the yaw angle and the pitch angle included in the posture information estimated in S104. By doing so, it is possible to correct the geometric deformation parameter while preventing the values of the yaw angle and pitch angle estimated in S104 from being greatly changed by the optimization process.
- the image processing apparatus 10 further includes an internal parameter acquisition unit that acquires already calibrated internal parameters, and the obtained five internal parameters and the yaw angle and pitch angle estimated by the posture estimation unit 110. And the remaining four parameters may be optimized.
- the five internal parameters are calibrated, for example, by performing camera calibration in advance in an imaging device (not shown) that captured the input image, and are acquired together with the input image.
- the image processing apparatus 10 can also accept a moving image as an input image, divide the moving image into continuous still images, and self-calibrate five internal parameters from various postures of the same person in each still image. it can.
- the image processing apparatus 10 acquires the five internal parameters calibrated in this way, and sets the remaining four parameters with the seven parameters combined with the yaw angle and the pitch angle estimated by the posture estimation unit 110 as fixed values. Optimize the parameters.
- the image processing apparatus 10 generates a normalized face image using the corrected geometric deformation parameter (S1064). Specifically, the image processing apparatus 10 converts each feature point on the three-dimensional solid shape model to the coordinates of the input image based on the corrected geometric deformation parameter and the coordinates of each feature point on the three-dimensional solid shape model. Calculate the coordinates when projected onto the system. Then, the image processing apparatus 10 backprojects the pixel corresponding to the calculated coordinates onto the three-dimensional solid shape model using the corrected geometric deformation parameter. Then, the image processing apparatus 10 uses the three-dimensional solid shape model in which the pixels are back-projected, and the normalized face in which the face position and size are constant and the face direction of the person is corrected to the front An image can be generated.
- the image processing apparatus 10 collates the normalized face image generated in S106 with the collation data (S202). Specifically, the image processing apparatus 10 extracts feature amounts from the normalized face image generated in S106.
- the feature amount extracted here is arbitrary. For example, a Gabor feature amount extracted by using Gabor Wavelet having a plurality of frequencies and angles may be used. The extracted feature amount is expressed by a vector, for example.
- the image processing apparatus 10 collates the feature amount vector extracted from the normalized face image with the feature amount vector of each matching data stored in the matching data storage unit 220, and calculates the matching score. To do. For example, normalized correlation or Euclidean distance can be used for matching feature quantity vectors. Further, the image processing apparatus 10 converts the feature quantity vector extracted in the feature quantity conversion matrix generated in advance by learning into a feature quantity vector having a lower dimension and excellent discrimination performance, and You may make it collate a normalized face image using the converted feature-value vector.
- the image processing apparatus 10 outputs the collation result of S202 to, for example, a display (S204).
- the face of the person in the input image is collated using the face image generated by the image processing apparatus 10.
- the precision of face authentication processing can be improved.
- the remaining parameters are optimized so that the square sum of the reprojection error is minimized while some of the geometric deformation parameters are fixed.
- the precision of optimization of a geometric deformation parameter can be improved. Therefore, the normalized face image can be generated with higher accuracy. Further, the calculation cost can be reduced by reducing the number of parameters to be optimized.
- FIG. 7 is a block diagram showing a detailed processing configuration of the image conversion unit 120 in the third embodiment.
- the image conversion unit 120 of the present embodiment further includes a weight coefficient calculation unit 128 in addition to the processing configuration of the image conversion unit 120 of the second embodiment.
- the weight coefficient calculation unit 128 calculates the contribution rate (weight coefficient) given to each of the plurality of feature points using the posture information of the face of the person in the input image estimated by the posture estimation unit 110. Specifically, according to the orientation of the person's face, the distance between each feature point of the person's face and a predetermined reference point such as the center of the lens of the imaging device differs. Therefore, the weighting factor calculation unit 128 sets a larger weighting factor for a feature point closer to a predetermined reference point, and sets a smaller weighting factor for a feature point farther from the predetermined reference point.
- the feature point on the left half of the face will be located on the near side of the feature point on the right half of the face, and the feature on the left half of the face
- the weighting coefficient assigned to the point is large, and the weighting coefficient assigned to the feature point on the right half of the face is small.
- the weighting factor calculation unit 128 sets the weighting factor of each feature point as follows. First, the weight coefficient calculation unit 128 rotates the three-dimensional face shape model based on the posture information estimated by the posture estimation unit 110 so that the posture is similar to the face of the person in the input image. Thereby, the weight coefficient calculation unit 128 can obtain the depth information of each feature point on the rotated three-dimensional solid shape model.
- the “depth information” is information indicating the depth from a predetermined reference point of each feature point on the three-dimensional solid shape model. Based on the depth information for each feature point, the weighting factor calculation unit 128 sets a larger weighting factor for a feature point that is closer to a predetermined reference point.
- the parameter calculation unit 122 of this embodiment calculates a geometric deformation parameter based on the weighting factor for each feature point calculated by the weighting factor calculation unit 128. Further, the parameter correction unit 124 of the present embodiment corrects the geometric deformation parameter based on the weighting factor for each feature point calculated by the weighting factor calculation unit 128.
- FIG. 8 is a flowchart showing a detailed flow of the process of S106 in the third embodiment.
- the image processing apparatus 10 calculates a weighting coefficient using the posture information estimated by the posture estimation unit 110 (S1065). Specifically, the image processing apparatus 10 rotates the three-dimensional solid shape model using the posture information estimated by the posture estimation unit 110, and each feature point of the rotated three-dimensional solid shape model and a predetermined reference point A weighting coefficient is assigned to each feature point according to the distance to
- the image processing apparatus 10 further uses the weighting factor for each feature point calculated in S1065 to calculate a geometric deformation parameter (S1062). Specifically, the image processing apparatus 10 calculates the geometric deformation parameter as follows. First, Expression 2 can be expressed as Expression 4 below.
- a diagonal matrix having a diagonal component as a weighting factor assigned to each feature point is assumed to be W. Then, when the pseudo inverse matrix calculation is performed in consideration of the weighting coefficient assigned to each feature point in Expression 4, Expression 5 is obtained.
- the image processing apparatus 10 can obtain a perspective projection transformation matrix in consideration of the weighting factor assigned to each feature amount based on Expression 5.
- the geometric deformation parameter can be calculated from the perspective projection transformation matrix.
- the image processing apparatus 10 further corrects the geometric deformation parameter by further using the weighting factor for each feature point calculated in S1065 (S1063). Specifically, since the reprojection error is calculated for each feature point, the image processing apparatus 10 calculates the reprojection error of each feature point in consideration of the weighting factor corresponding to each feature point. Then, as in the second embodiment, the image processing apparatus 10 corrects the geometric deformation parameter so that the square sum of the reprojection error is minimized.
- Posture estimation means for estimating posture information including a yaw angle and a pitch angle of the person's face from an input image including the person's face; The position of a plurality of feature points in a face area image that is an area including the face of the person in the input image, the position of the plurality of feature points in a three-dimensional solid model of the face of the person, and the posture information
- An image conversion means for generating a normalized face image in which the orientation of the face is corrected, An image processing apparatus.
- the image conversion means includes Geometric deformation that can convert between the coordinate system of the input image and the coordinate system of the three-dimensional solid shape model based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model Parameter calculating means for calculating the parameters; Parameter correction means for correcting the geometric deformation parameter based on the posture information; Normalized face image generation means for generating the normalized face image based on the corrected geometric deformation parameter; Having 1. An image processing apparatus according to 1. 3.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the parameter calculation means includes Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model;
- the parameter correction means includes Reprojection error of each feature point with initial values of parameters excluding the yaw angle and pitch angle of the geometric deformation parameters included in the perspective projection transformation matrix and the yaw angle and pitch angle included in the posture information Correcting the geometric deformation parameter included in the perspective projection transformation matrix so that the sum of squares of 2.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the parameter calculation means includes Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model;
- the parameter correction means includes Of the geometric deformation parameters included in the perspective projection transformation matrix, the yaw angle and pitch angle included in the posture information are fixed values, and the square sum of the reprojection error of each feature point is minimized. Correct the remaining nine parameters of the geometric deformation parameters; 2.
- An image processing apparatus according to 1. 5.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters,
- An internal parameter acquisition means for acquiring five calibrated internal parameters among the geometric deformation parameters;
- the parameter calculation means includes Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model;
- the parameter correction means includes Of the geometric deformation parameters included in the perspective projection transformation matrix, the calibrated five internal parameters and the yaw angle and pitch angle included in the posture information are fixed values, and the reprojection error of each feature point Correcting the remaining four parameters of the geometric deformation parameters so that the sum of squares of 2.
- An image processing apparatus according to 1. 6).
- the image conversion means includes A weight coefficient calculating means for calculating a weight coefficient to be given to each of the plurality of feature points based on the posture information;
- the parameter calculation means includes Further calculating the geometric deformation parameter using the weighting factor;
- the parameter correction means includes Correcting the geometric deformation parameter further using the weighting factor; 2.
- the image processing apparatus according to any one of the above. 7).
- the weight coefficient calculating means includes In the three-dimensional solid shape model rotated based on the posture information, depth information representing a depth from a predetermined reference point is acquired for each of the plurality of feature points, and the plurality of features is acquired based on the depth information. A feature point closer to the predetermined reference point among points is given a greater weight. 6).
- An image processing apparatus according to 1. 8).
- the image conversion means includes Generating the normalized face image in which the position, size, and orientation of the face are corrected to a constant state; 1. To 7. The image processing apparatus according to any one of the above. 9. Computer Estimating posture information including a yaw angle and a pitch angle of the person's face from an input image including the person's face, The position of a plurality of feature points in a face area image that is an area including the face of the person in the input image, the position of the plurality of feature points in a three-dimensional solid model of the face of the person, and the posture information To generate a normalized face image with the face orientation corrected, An image processing method. 10.
- the computer is Geometric deformation that can convert between the coordinate system of the input image and the coordinate system of the three-dimensional solid shape model based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model Calculate the parameters, Correcting the geometric deformation parameter based on the posture information; Generating the normalized face image based on the corrected geometric deformation parameter; Including An image processing method described in 1. 11.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the computer is Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model; Reprojection error of each feature point with initial values of parameters excluding the yaw angle and pitch angle of the geometric deformation parameters included in the perspective projection transformation matrix and the yaw angle and pitch angle included in the posture information Correcting the geometric deformation parameter included in the perspective projection transformation matrix so that the sum of squares of Including.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the computer is Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model; Of the geometric deformation parameters included in the perspective projection transformation matrix, the yaw angle and pitch angle included in the posture information are fixed values, and the square sum of the reprojection error of each feature point is minimized. Correct the remaining nine parameters of the geometric deformation parameters; Including. An image processing method described in 1. 13.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the computer is Among the geometric deformation parameters, obtain five calibrated internal parameters, Estimating a perspective projection transformation matrix including the geometric deformation parameter based on the correspondence relationship between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model; Of the geometric deformation parameters included in the perspective projection transformation matrix, the calibrated five internal parameters and the yaw angle and pitch angle included in the posture information are fixed values, and the reprojection error of each feature point Correcting the remaining four parameters of the geometric deformation parameters so that the sum of squares of Including. An image processing method described in 1.
- the computer is Based on the posture information, calculate a weighting factor to be given to each of the plurality of feature points, Further calculating the geometric deformation parameter using the weighting factor; Correcting the geometric deformation parameter further using the weighting factor; Including. To 13.
- the computer is In the three-dimensional solid shape model rotated based on the posture information, depth information representing a depth from a predetermined reference point is acquired for each of the plurality of feature points, and the plurality of features is acquired based on the depth information. A feature point closer to the predetermined reference point among points is given a greater weight. Including. An image processing method described in 1. 16.
- the computer is Generating the normalized face image in which the position, size, and orientation of the face are corrected to a constant state; Including To 15.
- the image processing method according to any one of the above. 17.
- Computer Posture estimation means for estimating posture information including a yaw angle and a pitch angle of the human face from an input image including the human face; The position of a plurality of feature points in a face area image that is an area including the face of the person in the input image, the position of the plurality of feature points in a three-dimensional solid model of the face of the person, and the posture information
- An image conversion means for generating a normalized face image in which the orientation of the face is corrected, Program to function as. 18.
- the computer In the image conversion means, Geometric deformation that can convert between the coordinate system of the input image and the coordinate system of the three-dimensional solid shape model based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model Parameter calculation means for calculating the parameters; Parameter correction means for correcting the geometric deformation parameter based on the posture information; Normalized face image generation means for generating the normalized face image based on the corrected geometric deformation parameter; 17. to function as The program described in. 19.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- In the parameter calculation means Based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model, a perspective projection transformation matrix including the geometric deformation parameter is estimated, In the parameter correction means, Reprojection error of each feature point with initial values of parameters excluding the yaw angle and pitch angle of the geometric deformation parameters included in the perspective projection transformation matrix and the yaw angle and pitch angle included in the posture information Correcting the geometric deformation parameter included in the perspective projection transformation matrix so that the square sum of 18.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- In the parameter calculation means Based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model, a perspective projection transformation matrix including the geometric deformation parameter is estimated, In the parameter correction means, Of the geometric deformation parameters included in the perspective projection transformation matrix, the yaw angle and pitch angle included in the posture information are fixed values, and the square sum of the reprojection error of each feature point is minimized. Correct the remaining nine parameters of the geometric deformation parameters; 18. The program described in. 21.
- the geometric deformation parameter has 5 internal parameters and 6 external parameters
- the computer Among the geometric deformation parameters, further function as an internal parameter acquisition means for acquiring five calibrated internal parameters, In the computer, In the parameter calculation means, Based on the correspondence between the positions of the plurality of feature points in the face area image and the three-dimensional solid shape model, a perspective projection transformation matrix including the geometric deformation parameter is estimated, In the parameter correction means, Of the geometric deformation parameters included in the perspective projection transformation matrix, the calibrated five internal parameters and the yaw angle and pitch angle included in the posture information are fixed values, and the reprojection error of each feature point Correcting the remaining four parameters of the geometric deformation parameters so that the sum of squares of 18. The program described in. 22.
- the computer In the image conversion means, Based on the posture information, further function as a weighting factor calculating means for calculating a weighting factor to be given to each of the plurality of feature points, In the computer, In the parameter calculation means, Further calculating the geometric deformation parameter using the weighting factor; In the parameter correction means, Correcting the geometric deformation parameter further using the weighting factor; 18. To 21. The program as described in any one of these. 23. In the computer, In the weight coefficient calculating means, In the three-dimensional solid shape model rotated based on the posture information, depth information representing a depth from a predetermined reference point is acquired for each of the plurality of feature points, and the plurality of features is acquired based on the depth information.
- a feature point closer to the predetermined reference point among points is given a greater weight. 22.
- the image conversion means Generating the normalized face image in which the position, size, and orientation of the face are corrected to a constant state; 17.
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Description
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段と、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段と、
を有する画像処理装置が提供される。
コンピュータが、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定し、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する、
ことを含む画像処理方法が提供される。
コンピュータを、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段、
として機能させるためのプログラムが提供される。
図1は、第1実施形態における画像処理装置10の構成例を示すブロック図である。画像処理装置10は姿勢推定部110及び画像変換部120を有する。
・安藤、草地 他、「サポートベクトル回帰を用いた三次元物体の姿勢推定法」、電子情報通信学会論文誌
・山田、中島 他「因子分解法と部分空間法による顔向き推定」、電子情報通信学会技術研究報告PRMU
・特開2011-209116号公報
・佐藤洋一、「コンピュータビジョンによる顔のトラッキング」、映像情報メディア学会誌
ただし、上述した姿勢情報を推定する手法はあくまで一例であり、姿勢推定部110は、その他の既知の手法を用いて人物の顔の姿勢情報を推定することもできる。
本実施形態では、画像変換部120の詳細な処理構成について説明する。また、本実施形態の画像処理装置10は顔画像を照合するための構成を更に含む。
本実施形態は、以下の点を除いて、第1及び第2実施形態と同様である。
1.人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段と、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段と、
を有する画像処理装置。
2.前記画像変換手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記入力画像の座標系と前記3次元立体形状モデルの座標系とを変換可能な幾何変形パラメータを算出するパラメータ算出手段と、
前記姿勢情報に基づいて前記幾何変形パラメータを補正するパラメータ補正手段と、
補正された前記幾何変形パラメータに基づいて、前記正規化顔画像を生成する正規化顔画像生成手段と、
を有する1.に記載の画像処理装置。
3.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうちのヨー角及びピッチ角を除くパラメータと、前記姿勢情報に含まれるヨー角及びピッチ角とを初期値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記透視投影変換行列に含まれる前記幾何変形パラメータを補正する、
2.に記載の画像処理装置。
4.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記姿勢情報に含まれるヨー角及びピッチ角を固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの9個のパラメータを補正する、
2.に記載の画像処理装置。
5.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記幾何変形パラメータのうち、校正された5個の内部パラメータを取得する内部パラメータ取得手段を更に備え、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記校正された5個の内部パラメータと前記姿勢情報に含まれるヨー角及びピッチ角とを固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの4個のパラメータを補正する、
2.に記載の画像処理装置。
6.前記画像変換手段は、
前記姿勢情報に基づいて、前記複数の特徴点毎に与える重み係数を算出する重み係数算出手段を更に備え、
前記パラメータ算出手段は、
前記重み係数を更に用いて前記幾何変形パラメータを算出し、
前記パラメータ補正手段は、
前記重み係数を更に用いて前記幾何変形パラメータを補正する、
2.から5.のいずれか1つに記載の画像処理装置。
7.前記重み係数算出手段は、
前記姿勢情報に基づいて回転させた前記3次元立体形状モデルにおいて、所定の基準点からの奥行きを表す奥行き情報を前記複数の特徴点毎に取得し、前記奥行き情報に基づいて、前記複数の特徴点のうち前記所定の基準点に近い特徴点ほど大きい重みを与える、
6.に記載の画像処理装置。
8.前記画像変換手段は、
顔の位置、大きさ、及び向きが一定の状態に補正された前記正規化顔画像を生成する、
1.から7.のいずれか1つに記載の画像処理装置。
9.コンピュータが、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定し、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する、
ことを含む画像処理方法。
10.前記コンピュータが、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記入力画像の座標系と前記3次元立体形状モデルの座標系とを変換可能な幾何変形パラメータを算出し、
前記姿勢情報に基づいて前記幾何変形パラメータを補正し、
補正された前記幾何変形パラメータに基づいて、前記正規化顔画像を生成する、
ことを含む9.に記載の画像処理方法。
11.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータが、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうちのヨー角及びピッチ角を除くパラメータと、前記姿勢情報に含まれるヨー角及びピッチ角とを初期値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記透視投影変換行列に含まれる前記幾何変形パラメータを補正する、
ことを含む10.に記載の画像処理方法。
12.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータが、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記姿勢情報に含まれるヨー角及びピッチ角を固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの9個のパラメータを補正する、
ことを含む10.に記載の画像処理方法。
13.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータが、
前記幾何変形パラメータのうち、校正された5個の内部パラメータを取得し、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記校正された5個の内部パラメータと前記姿勢情報に含まれるヨー角及びピッチ角とを固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの4個のパラメータを補正する、
ことを含む10.に記載の画像処理方法。
14.前記コンピュータが、
前記姿勢情報に基づいて、前記複数の特徴点毎に与える重み係数を算出し、
前記重み係数を更に用いて前記幾何変形パラメータを算出し、
前記重み係数を更に用いて前記幾何変形パラメータを補正する、
ことを含む10.から13.のいずれか1つに記載の画像処理方法。
15.前記コンピュータが、
前記姿勢情報に基づいて回転させた前記3次元立体形状モデルにおいて、所定の基準点からの奥行きを表す奥行き情報を前記複数の特徴点毎に取得し、前記奥行き情報に基づいて、前記複数の特徴点のうち前記所定の基準点に近い特徴点ほど大きい重みを与える、
ことを含む14.に記載の画像処理方法。
16.前記コンピュータが、
顔の位置、大きさ、及び向きが一定の状態に補正された前記正規化顔画像を生成する、
ことを含む9.から15.のいずれか1つに記載の画像処理方法。
17. コンピュータを、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段、
として機能させるためのプログラム。
18.前記コンピュータを、
前記画像変換手段において、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記入力画像の座標系と前記3次元立体形状モデルの座標系とを変換可能な幾何変形パラメータを算出するパラメータ算出手段、
前記姿勢情報に基づいて前記幾何変形パラメータを補正するパラメータ補正手段と、
補正された前記幾何変形パラメータに基づいて、前記正規化顔画像を生成する正規化顔画像生成手段、
として機能させるための17.に記載のプログラム。
19.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータに、
前記パラメータ算出手段において、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定させ、
前記パラメータ補正手段において、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうちのヨー角及びピッチ角を除くパラメータと、前記姿勢情報に含まれるヨー角及びピッチ角とを初期値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記透視投影変換行列に含まれる前記幾何変形パラメータを補正させる、
18.に記載のプログラム。
20.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータに、
前記パラメータ算出手段において、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定させ、
前記パラメータ補正手段において、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記姿勢情報に含まれるヨー角及びピッチ角を固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの9個のパラメータを補正させる、
18.に記載のプログラム。
21.前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記コンピュータを、
前記幾何変形パラメータのうち、校正された5個の内部パラメータを取得する内部パラメータ取得手段として更に機能させ、
前記コンピュータに、
前記パラメータ算出手段において、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定させ、
前記パラメータ補正手段において、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記校正された5個の内部パラメータと前記姿勢情報に含まれるヨー角及びピッチ角とを固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの4個のパラメータを補正させる、
18.に記載のプログラム。
22.前記コンピュータを、
前記画像変換手段において、
前記姿勢情報に基づいて、前記複数の特徴点毎に与える重み係数を算出する重み係数算出手段として更に機能させ、
前記コンピュータに、
前記パラメータ算出手段において、
前記重み係数を更に用いて前記幾何変形パラメータを算出させ、
前記パラメータ補正手段において、
前記重み係数を更に用いて前記幾何変形パラメータを補正させる、
18.から21.のいずれか1つに記載のプログラム。
23.前記コンピュータに、
前記重み係数算出手段において、
前記姿勢情報に基づいて回転させた前記3次元立体形状モデルにおいて、所定の基準点からの奥行きを表す奥行き情報を前記複数の特徴点毎に取得させ、前記奥行き情報に基づいて、前記複数の特徴点のうち前記所定の基準点に近い特徴点ほど大きい重みを与えさせる、
22.に記載のプログラム。
24.前記コンピュータに、
前記画像変換手段において、
顔の位置、大きさ、及び向きが一定の状態に補正された前記正規化顔画像を生成させる、
17.から23.のいずれか1つに記載のプログラム。
Claims (10)
- 人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段と、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段と、
を有する画像処理装置。 - 前記画像変換手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記入力画像の座標系と前記3次元立体形状モデルの座標系とを変換可能な幾何変形パラメータを算出するパラメータ算出手段と、
前記姿勢情報に基づいて前記幾何変形パラメータを補正するパラメータ補正手段と、
補正された前記幾何変形パラメータに基づいて、前記正規化顔画像を生成する正規化顔画像生成手段と、
を有する請求項1に記載の画像処理装置。 - 前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうちのヨー角及びピッチ角を除くパラメータと、前記姿勢情報に含まれるヨー角及びピッチ角とを初期値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記透視投影変換行列に含まれる前記幾何変形パラメータを補正する、
請求項2に記載の画像処理装置。 - 前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記姿勢情報に含まれるヨー角及びピッチ角を固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの9個のパラメータを補正する、
請求項2に記載の画像処理装置。 - 前記幾何変形パラメータは、5個の内部パラメータと6個の外部パラメータを有しており、
前記幾何変形パラメータのうち、校正された5個の内部パラメータを取得する内部パラメータ取得手段を更に備え、
前記パラメータ算出手段は、
前記顔領域画像と前記3次元立体形状モデルとにおける前記複数の特徴点の位置の対応関係に基づいて、前記幾何変形パラメータを含む透視投影変換行列を推定し、
前記パラメータ補正手段は、
前記透視投影変換行列に含まれる前記幾何変形パラメータのうち、前記校正された5個の内部パラメータと前記姿勢情報に含まれるヨー角及びピッチ角とを固定値として、前記各特徴点の再投影誤差の2乗和が最小となるように前記幾何変形パラメータの残りの4個のパラメータを補正する、
請求項2に記載の画像処理装置。 - 前記画像変換手段は、
前記姿勢情報に基づいて、前記複数の特徴点毎に与える重み係数を算出する重み係数算出手段を更に備え、
前記パラメータ算出手段は、
前記重み係数を更に用いて前記幾何変形パラメータを算出し、
前記パラメータ補正手段は、
前記重み係数を更に用いて前記幾何変形パラメータを補正する、
請求項2から5のいずれか1項に記載の画像処理装置。 - 前記重み係数算出手段は、
前記姿勢情報に基づいて回転させた前記3次元立体形状モデルにおいて、所定の基準点からの奥行きを表す奥行き情報を前記複数の特徴点毎に取得し、前記奥行き情報に基づいて、前記複数の特徴点のうち前記所定の基準点に近い特徴点ほど大きい重みを与える、
請求項6に記載の画像処理装置。 - 前記画像変換手段は、
顔の位置、大きさ、及び向きが一定の状態に補正された前記正規化顔画像を生成する、
請求項1から7のいずれか1項に記載の画像処理装置。 - コンピュータが、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定し、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する、
ことを含む画像処理方法。 - コンピュータを、
人物の顔を含む入力画像から、前記人物の顔のヨー角及びピッチ角を含む姿勢情報を推定する姿勢推定手段、
前記入力画像中の前記人物の顔を含む領域である顔領域画像における複数の特徴点の位置と、人物の顔の3次元立体形状モデルにおける前記複数の特徴点の位置と、前記姿勢情報とに基づいて、顔の向きが補正された正規化顔画像を生成する画像変換手段、
として機能させるためのプログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US14/914,321 US9881203B2 (en) | 2013-08-29 | 2014-08-26 | Image processing device, image processing method, and program |
JP2015534223A JP6424822B2 (ja) | 2013-08-29 | 2014-08-26 | 画像処理装置、画像処理方法、及びプログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013-178461 | 2013-08-29 | ||
JP2013178461 | 2013-08-29 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015029982A1 true WO2015029982A1 (ja) | 2015-03-05 |
Family
ID=52586539
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2014/072258 WO2015029982A1 (ja) | 2013-08-29 | 2014-08-26 | 画像処理装置、画像処理方法、及びプログラム |
Country Status (3)
Country | Link |
---|---|
US (1) | US9881203B2 (ja) |
JP (1) | JP6424822B2 (ja) |
WO (1) | WO2015029982A1 (ja) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016181051A (ja) * | 2015-03-23 | 2016-10-13 | キヤノン株式会社 | 画像処理装置、画像処理方法 |
CN106295496A (zh) * | 2015-06-24 | 2017-01-04 | 三星电子株式会社 | 脸部识别方法和设备 |
JP2017083230A (ja) * | 2015-10-26 | 2017-05-18 | トヨタ自動車株式会社 | 自己位置推定方法 |
JP6431591B1 (ja) * | 2017-12-15 | 2018-11-28 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP6490861B1 (ja) * | 2018-08-17 | 2019-03-27 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP2019096113A (ja) * | 2017-11-24 | 2019-06-20 | Kddi株式会社 | キーポイントデータに関する加工装置、方法及びプログラム |
CN110427849A (zh) * | 2019-07-23 | 2019-11-08 | 深圳前海达闼云端智能科技有限公司 | 人脸姿态确定方法、装置、存储介质和电子设备 |
US10733422B2 (en) | 2015-06-24 | 2020-08-04 | Samsung Electronics Co., Ltd. | Face recognition method and apparatus |
JP2020144546A (ja) * | 2019-03-05 | 2020-09-10 | アルパイン株式会社 | 顔特徴部検出装置、顔特徴部検出方法 |
WO2020195810A1 (ja) * | 2019-03-26 | 2020-10-01 | 日本電信電話株式会社 | 重みマスク生成装置、重みマスク生成方法及びプログラム |
CN111985280A (zh) * | 2019-05-24 | 2020-11-24 | 北京小米移动软件有限公司 | 图像处理方法及装置 |
CN112949576A (zh) * | 2021-03-29 | 2021-06-11 | 北京京东方技术开发有限公司 | 姿态估计方法、装置、设备及存储介质 |
JP2021117049A (ja) * | 2020-01-23 | 2021-08-10 | 国立大学法人電気通信大学 | 口腔認識装置 |
Families Citing this family (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106033621B (zh) * | 2015-03-17 | 2018-08-24 | 阿里巴巴集团控股有限公司 | 一种三维建模的方法及装置 |
JP6536169B2 (ja) * | 2015-05-21 | 2019-07-03 | カシオ計算機株式会社 | 画像処理装置、画像処理方法及びプログラム |
US9986150B2 (en) * | 2015-09-30 | 2018-05-29 | Ricoh Co., Ltd. | Algorithm to estimate yaw errors in camera pose |
US10104282B2 (en) | 2015-09-30 | 2018-10-16 | Ricoh Co., Ltd. | Yaw user interface |
EP3232368A1 (en) * | 2016-04-14 | 2017-10-18 | Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. | Determining facial parameters |
US9961261B2 (en) * | 2016-06-20 | 2018-05-01 | Gopro, Inc. | Image alignment using a virtual gyroscope model |
WO2018031946A1 (en) * | 2016-08-11 | 2018-02-15 | MetaMason, Inc. | Customized cpap masks and related modeling algorithms |
JP7071054B2 (ja) * | 2017-01-20 | 2022-05-18 | キヤノン株式会社 | 情報処理装置、情報処理方法およびプログラム |
US10158797B2 (en) * | 2017-03-31 | 2018-12-18 | Motorola Mobility Llc | Combining images when a face is present |
CN108960020A (zh) * | 2017-05-27 | 2018-12-07 | 富士通株式会社 | 信息处理方法和信息处理设备 |
JP6773061B2 (ja) | 2018-02-16 | 2020-10-21 | 新東工業株式会社 | 評価システム、評価装置、評価方法、評価プログラム、及び記録媒体 |
US10706577B2 (en) * | 2018-03-06 | 2020-07-07 | Fotonation Limited | Facial features tracker with advanced training for natural rendering of human faces in real-time |
EP3537447A1 (en) * | 2018-03-07 | 2019-09-11 | Koninklijke Philips N.V. | Display of medical image data |
US11282543B2 (en) * | 2018-03-09 | 2022-03-22 | Apple Inc. | Real-time face and object manipulation |
AU2019208182B2 (en) | 2018-07-25 | 2021-04-08 | Konami Gaming, Inc. | Casino management system with a patron facial recognition system and methods of operating same |
US11521460B2 (en) | 2018-07-25 | 2022-12-06 | Konami Gaming, Inc. | Casino management system with a patron facial recognition system and methods of operating same |
JP6863946B2 (ja) * | 2018-10-31 | 2021-04-21 | ファナック株式会社 | 画像処理装置 |
CN111178117B (zh) * | 2018-11-09 | 2023-09-22 | 浙江宇视科技有限公司 | 数据处理方法及装置 |
CN109376684B (zh) * | 2018-11-13 | 2021-04-06 | 广州市百果园信息技术有限公司 | 一种人脸关键点检测方法、装置、计算机设备和存储介质 |
CN111288956B (zh) * | 2018-12-07 | 2022-04-22 | 顺丰科技有限公司 | 一种目标姿态确定方法、装置、设备和存储介质 |
CN111414798B (zh) * | 2019-02-03 | 2022-12-06 | 沈阳工业大学 | 基于rgb-d图像的头部姿态检测方法及*** |
US11610414B1 (en) * | 2019-03-04 | 2023-03-21 | Apple Inc. | Temporal and geometric consistency in physical setting understanding |
TWI720447B (zh) | 2019-03-28 | 2021-03-01 | 財團法人工業技術研究院 | 影像定位方法及其系統 |
CN111986097B (zh) * | 2019-05-24 | 2024-02-09 | 北京小米移动软件有限公司 | 图像处理方法及装置 |
CN112529784B (zh) * | 2019-09-18 | 2024-05-28 | 华为技术有限公司 | 图像畸变校正方法及装置 |
CN111640055B (zh) * | 2020-05-22 | 2023-04-11 | 构范(厦门)信息技术有限公司 | 一种二维人脸图片变形方法及*** |
CN112562048A (zh) * | 2020-12-16 | 2021-03-26 | 北京百度网讯科技有限公司 | 三维模型的控制方法、装置、设备以及存储介质 |
CN113011401B (zh) * | 2021-04-30 | 2023-03-21 | 汇纳科技股份有限公司 | 人脸图像姿态估计和校正方法、***、介质及电子设备 |
CN113362231B (zh) * | 2021-07-23 | 2024-06-25 | 百果园技术(新加坡)有限公司 | 人脸关键点的插值方法、装置、计算机设备和存储介质 |
CN113936324A (zh) * | 2021-10-29 | 2022-01-14 | Oppo广东移动通信有限公司 | 注视检测方法、电子设备的控制方法及相关设备 |
CN117135443A (zh) * | 2023-02-22 | 2023-11-28 | 荣耀终端有限公司 | 一种图像抓拍方法及电子设备 |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009020761A (ja) * | 2007-07-12 | 2009-01-29 | Toshiba Corp | 画像処理装置及びその方法 |
JP2009053916A (ja) * | 2007-08-27 | 2009-03-12 | Sony Corp | 顔画像処理装置及び顔画像処理方法、並びにコンピュータ・プログラム |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005339288A (ja) * | 2004-05-27 | 2005-12-08 | Toshiba Corp | 画像処理装置及びその方法 |
JP2004288222A (ja) | 2004-07-13 | 2004-10-14 | Nec Corp | 画像照合装置及びその画像照合方法並びにその制御プログラムを記録した記録媒体 |
JP5018029B2 (ja) | 2006-11-10 | 2012-09-05 | コニカミノルタホールディングス株式会社 | 認証システム及び認証方法 |
JP4389956B2 (ja) * | 2007-04-04 | 2009-12-24 | ソニー株式会社 | 顔認識装置及び顔認識方法、並びにコンピュータ・プログラム |
JP4577410B2 (ja) * | 2008-06-18 | 2010-11-10 | ソニー株式会社 | 画像処理装置、画像処理方法およびプログラム |
JP2013022705A (ja) * | 2011-07-25 | 2013-02-04 | Sony Corp | ロボット装置及びロボット装置の制御方法、コンピューター・プログラム、並びにロボット・システム |
-
2014
- 2014-08-26 JP JP2015534223A patent/JP6424822B2/ja active Active
- 2014-08-26 US US14/914,321 patent/US9881203B2/en active Active
- 2014-08-26 WO PCT/JP2014/072258 patent/WO2015029982A1/ja active Application Filing
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009020761A (ja) * | 2007-07-12 | 2009-01-29 | Toshiba Corp | 画像処理装置及びその方法 |
JP2009053916A (ja) * | 2007-08-27 | 2009-03-12 | Sony Corp | 顔画像処理装置及び顔画像処理方法、並びにコンピュータ・プログラム |
Non-Patent Citations (1)
Title |
---|
TATSUO KOSAKATANI ET AL.: "Projection-based 3D Normalization for Face Recognition", IEICE TECHNICAL REPORT, vol. 105, no. 375, 21 October 2005 (2005-10-21), pages 49 - 54 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016181051A (ja) * | 2015-03-23 | 2016-10-13 | キヤノン株式会社 | 画像処理装置、画像処理方法 |
US10733422B2 (en) | 2015-06-24 | 2020-08-04 | Samsung Electronics Co., Ltd. | Face recognition method and apparatus |
CN106295496A (zh) * | 2015-06-24 | 2017-01-04 | 三星电子株式会社 | 脸部识别方法和设备 |
US11386701B2 (en) | 2015-06-24 | 2022-07-12 | Samsung Electronics Co., Ltd. | Face recognition method and apparatus |
CN106295496B (zh) * | 2015-06-24 | 2021-09-14 | 三星电子株式会社 | 脸部识别方法和设备 |
JP2017083230A (ja) * | 2015-10-26 | 2017-05-18 | トヨタ自動車株式会社 | 自己位置推定方法 |
JP7015152B2 (ja) | 2017-11-24 | 2022-02-15 | Kddi株式会社 | キーポイントデータに関する加工装置、方法及びプログラム |
JP2019096113A (ja) * | 2017-11-24 | 2019-06-20 | Kddi株式会社 | キーポイントデータに関する加工装置、方法及びプログラム |
JP2019109313A (ja) * | 2017-12-15 | 2019-07-04 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP6431591B1 (ja) * | 2017-12-15 | 2018-11-28 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP6490861B1 (ja) * | 2018-08-17 | 2019-03-27 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP2019109491A (ja) * | 2018-08-17 | 2019-07-04 | 株式会社シャルマン | 三次元顔画像の基準正面の設定方法、それを用いた眼鏡の選定方法及びそれを用いたカルテの作成方法 |
JP2020144546A (ja) * | 2019-03-05 | 2020-09-10 | アルパイン株式会社 | 顔特徴部検出装置、顔特徴部検出方法 |
JP7055762B2 (ja) | 2019-03-05 | 2022-04-18 | アルパイン株式会社 | 顔特徴部検出装置、顔特徴部検出方法 |
WO2020195810A1 (ja) * | 2019-03-26 | 2020-10-01 | 日本電信電話株式会社 | 重みマスク生成装置、重みマスク生成方法及びプログラム |
JP2020160792A (ja) * | 2019-03-26 | 2020-10-01 | 日本電信電話株式会社 | 重みマスク生成装置、重みマスク生成方法及びプログラム |
JP7161111B2 (ja) | 2019-03-26 | 2022-10-26 | 日本電信電話株式会社 | 重みマスク生成装置、重みマスク生成方法及びプログラム |
CN111985280A (zh) * | 2019-05-24 | 2020-11-24 | 北京小米移动软件有限公司 | 图像处理方法及装置 |
CN111985280B (zh) * | 2019-05-24 | 2023-12-29 | 北京小米移动软件有限公司 | 图像处理方法及装置 |
CN110427849B (zh) * | 2019-07-23 | 2022-02-08 | 达闼机器人有限公司 | 人脸姿态确定方法、装置、存储介质和电子设备 |
CN110427849A (zh) * | 2019-07-23 | 2019-11-08 | 深圳前海达闼云端智能科技有限公司 | 人脸姿态确定方法、装置、存储介质和电子设备 |
JP2021117049A (ja) * | 2020-01-23 | 2021-08-10 | 国立大学法人電気通信大学 | 口腔認識装置 |
JP7414262B2 (ja) | 2020-01-23 | 2024-01-16 | 国立大学法人電気通信大学 | 口腔認識装置 |
CN112949576A (zh) * | 2021-03-29 | 2021-06-11 | 北京京东方技术开发有限公司 | 姿态估计方法、装置、设备及存储介质 |
CN112949576B (zh) * | 2021-03-29 | 2024-04-23 | 北京京东方技术开发有限公司 | 姿态估计方法、装置、设备及存储介质 |
Also Published As
Publication number | Publication date |
---|---|
US9881203B2 (en) | 2018-01-30 |
US20160217318A1 (en) | 2016-07-28 |
JP6424822B2 (ja) | 2018-11-21 |
JPWO2015029982A1 (ja) | 2017-03-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6424822B2 (ja) | 画像処理装置、画像処理方法、及びプログラム | |
US8055028B2 (en) | Object pose normalization method and apparatus and object recognition method | |
JP4728432B2 (ja) | 顔姿勢推定装置、顔姿勢推定方法、及び、顔姿勢推定プログラム | |
US9619704B2 (en) | Fast articulated motion tracking | |
US8155398B2 (en) | Image processing apparatus, image processing method and computer program | |
US8208717B2 (en) | Combining subcomponent models for object image modeling | |
US8515180B2 (en) | Image data correction apparatus and method using feature points vector data | |
US8131063B2 (en) | Model-based object image processing | |
JP4469275B2 (ja) | ポーズ角度を推定するためのシステムおよび方法 | |
US8204301B2 (en) | Iterative data reweighting for balanced model learning | |
US7894636B2 (en) | Apparatus and method for performing facial recognition from arbitrary viewing angles by texturing a 3D model | |
JP4950787B2 (ja) | 画像処理装置及びその方法 | |
US8260039B2 (en) | Object model fitting using manifold constraints | |
EP1677250B1 (en) | Image collation system and image collation method | |
US8260038B2 (en) | Subdivision weighting for robust object model fitting | |
US9443325B2 (en) | Image processing apparatus, image processing method, and computer program | |
JP2007058393A (ja) | 認証装置、認証方法及びプログラム | |
JP4539519B2 (ja) | 立体モデル生成装置および立体モデル生成方法 | |
US10902628B1 (en) | Method for estimating user eye orientation using a system-independent learned mapping | |
JP6086491B2 (ja) | 画像処理装置およびそのデータベース構築装置 | |
JP6003367B2 (ja) | 画像認識装置、画像認識方法および画像認識プログラム | |
CN108694348B (zh) | 一种基于自然特征的跟踪注册方法及装置 | |
JP2007058402A (ja) | 認証システム、認証方法およびプログラム | |
US20210312648A1 (en) | Face image processing device and face image processing program | |
JP6260146B2 (ja) | 形状状態推定装置、形状モデル生成装置、及びプログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 14839803 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2015534223 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14914321 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 14839803 Country of ref document: EP Kind code of ref document: A1 |