CN109886183B - Human face age estimation method and device based on bridge type neural network - Google Patents
Human face age estimation method and device based on bridge type neural network Download PDFInfo
- Publication number
- CN109886183B CN109886183B CN201910121339.XA CN201910121339A CN109886183B CN 109886183 B CN109886183 B CN 109886183B CN 201910121339 A CN201910121339 A CN 201910121339A CN 109886183 B CN109886183 B CN 109886183B
- Authority
- CN
- China
- Prior art keywords
- gate
- local
- network
- face
- regressor
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 29
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 9
- 238000005516 engineering process Methods 0.000 claims abstract description 9
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 76
- 238000012549 training Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000001815 facial effect Effects 0.000 description 5
- 230000004913 activation Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009759 skin aging Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method and a device for estimating face age based on a bridge type neural network, wherein the method comprises the following steps: acquiring an input picture to be estimated; acquiring a face area and a plurality of key points of a face by a face detection technology according to an input picture; performing face alignment according to the plurality of key points to obtain an aligned face picture; and inputting the aligned face picture into a deep convolutional neural network for feature extraction, connecting the full-connection layer for feature extraction to a local regressor and a gate network of a bridge network to generate a regression result and a gate function, and weighting the regression result and the gate function to obtain an age estimation result. The method can solve the problem of low accuracy rate in the existing age estimation, and utilizes a bridge network for sensing continuity to enable the continuity relation between local regressors to be explicitly modeled, so that the information is fully utilized to improve the final age estimation performance and improve the accuracy rate.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a method and a device for estimating face age based on a bridge type neural network.
Background
The task of automatic estimation of the face age is to automatically estimate an accurate age value for a given face picture. The face age estimation has wide application in video monitoring, man-machine interaction, social media, face retrieval and the like. Although this problem has been studied for many years, it remains quite challenging to give an accurate age estimate.
The existing face recognition technology can be divided into three categories: regression-based methods, classification-based methods, and sequence-based methods. Regression-based methods consider the age label as a numerical value and then use a regressor to regress the age directly. However, faces mature in different ways at different ages, for example, changes in the face during childhood are mainly reflected in the growth of the skeleton and changes in the face during the old age are mainly reflected in skin aging. This non-stationary aging process means that the age estimated data is heterogeneous, so that a common global regressor has difficulty in handling heterogeneous data, all of which are easily over-fitted on such data. The divide and conquer method is proved to be a good method for processing heterogeneous data, and divides the whole data space into a plurality of subspaces and uses a regressor to regress each subspace. However, the aging process of the face is also a continuous process, that is, the face gradually changes with age. For example, the facial appearance of a person is similar between the ages of 31 and 32. Such a continuity is not fully exploited in the prior art methods. The classification-based approach treats different ages as independent age labels, so any type of classification error is the same, which ignores the correlation between age labels. Recently, many sequence-based methods have been proposed, which generally use a series of binary classifiers to determine the sequence of the age, and combine the results of these binary classifiers to obtain the final age estimation value. However, this type of approach also ignores the association between binary classifiers, resulting in limited performance.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, an object of the present invention is to provide a method for estimating a human face age based on a bridge neural network, which can solve the problem of low accuracy rate in the existing age estimation.
The invention also aims to provide a human face age estimation device based on the bridge type neural network.
In order to achieve the above object, an embodiment of the present invention provides a face age estimation method based on a bridge neural network, including: acquiring an input picture to be estimated; acquiring a face area and a plurality of key points of a face by a face detection technology according to the input picture; performing face alignment according to the plurality of key points to obtain an aligned face picture; and inputting the aligned face picture into a deep convolutional neural network for feature extraction, connecting the full-connection layer for feature extraction to a local regressor and a gate network of a bridge network to generate a regression result and a gate function, and weighting the regression result and the gate function to obtain an age estimation result.
According to the face age estimation method based on the bridge neural network, the bridge network for sensing continuity is provided, so that the continuity relation among the local regressors is explicitly modeled, the final age estimation performance is improved by fully utilizing the information, and the accuracy is improved.
In addition, the face age estimation method based on the bridge neural network according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the weighting the regression result and the gate function to obtain the age estimation result further includes:
dividing training data into k subsets with overlap by the local regressor to process heterogeneous data, training one local regressor for each subset, and recording the regression result value of the ith local regressor as ul(x) The gate network generates one gate function for each local regressor, and the gate function generated by the ith gate network is recorded as pil(x) The formula of the age estimation result is as follows:
where x is the input picture sample and y is the estimated age.
Further, in one embodiment of the present invention, the local regressors divide the entire data space into a plurality of subspaces, and each local regressor performs regression in one subspace.
Further, in one embodiment of the present invention, the local regressor generates a regression result value ul(x) And said gate network generates one said gate function pi for each said local regressorl(x) To perform weighting.
Further, in one embodiment of the invention, the gate function pi is generated in the gate network by a bridge tree structurel(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value on the edge from node m to node n.
In order to achieve the above object, another embodiment of the present invention provides a facial age estimation apparatus based on a bridge neural network, including: the first acquisition module is used for acquiring an input picture to be estimated; the second acquisition module is used for acquiring a face area and a plurality of key points of a face according to the input picture by a face detection technology; the alignment module is used for carrying out face alignment according to the plurality of key points so as to obtain an aligned face picture; and the estimation module is used for inputting the aligned face pictures into a deep convolutional neural network for feature extraction, connecting the full-connection layer for feature extraction to a local regressor and a gate network of a bridge network to generate a regression result and a gate function, and weighting and summing the regression result and the gate function to obtain an age estimation result.
According to the face age estimation device based on the bridge neural network, provided by the embodiment of the invention, the continuity relation among local regressors is explicitly modeled by providing the bridge network for sensing continuity, so that the final age estimation performance is improved by fully utilizing the information, and the accuracy is improved.
In addition, the face age estimation device based on the bridge neural network according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the estimation module is further configured to:
dividing training data into k subsets with overlap by the local regressor to process heterogeneous data, training one local regressor for each subset, and recording the regression result value of the ith local regressor as ul(x) The gate network generates one gate function for each local regressor, and the gate function generated by the ith gate network is recorded as pil(x) The formula of the age estimation result is as follows:
where x is the input picture sample and y is the estimated age.
Further, in one embodiment of the present invention, the local regressors divide the entire data space into a plurality of subspaces, and each local regressor performs regression in one subspace.
Further, in one embodiment of the present invention, the local regressor generates a regression result value ul(x) And the door network corresponds to each of the door networksEach local regressor generates one gate function pil(x) To perform weighting.
Further, in one embodiment of the invention, the gate function pi is generated in the gate network by a bridge tree structurel(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value on the edge from node m to node n.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flowchart of a method for estimating a face age based on a bridge neural network according to an embodiment of the present invention;
FIG. 2 is a flowchart of a method for estimating a face age based on a bridge neural network according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a bridge tree structure according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating a bridge tree structure according to another embodiment of the present invention;
FIG. 5 is a detailed diagram of an implementation of a gate network according to an embodiment of the invention;
fig. 6 is a schematic structural diagram of a facial age estimation device based on a bridge neural network according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following describes a method and an apparatus for estimating a face age based on a bridge neural network according to an embodiment of the present invention with reference to the accompanying drawings.
First, a method for estimating a face age based on a bridge neural network according to an embodiment of the present invention will be described with reference to the accompanying drawings.
Fig. 1 is a flowchart of a face age estimation method based on a bridge neural network according to an embodiment of the present invention.
As shown in fig. 1, the method for estimating the age of a human face based on a bridge neural network includes the following steps:
in step S101, an input picture to be estimated is acquired.
In step S102, a face region and a plurality of key points of a face are obtained by a face detection technique according to an input picture.
Specifically, as shown in fig. 2, an input picture x of a face is obtained, and a face detection technology MTCNN is used to detect a face region and five key points of the face.
In step S103, face alignment is performed according to the plurality of key points to obtain an aligned face picture.
Specifically, the face alignment is performed according to the face key points acquired in step S102, and an aligned face picture is generated.
In step S104, the aligned face image is input into a deep convolutional neural network for feature extraction, the full link layer for feature extraction is connected to a local regressor and a gate network of a bridge network, a regression result and a gate function are generated, and the regression result and the gate function are weighted and summed to obtain an age estimation result.
Specifically, the aligned face picture is sent into a deep convolutional neural network for feature extraction, and then a full connection layer in the deep convolutional neural network for feature extraction is connected to two parts of a bridge network: a local regressor and a gate network, the final age estimate being obtained by a weighted sum of the local regressors.
The following details the two components of the bridge network (local regressor and gate network) and how to combine them to obtain the final age estimate.
Further, in an embodiment of the present invention, the weighting the regression result and the gate function to obtain the age estimation result further includes: dividing training data into k subsets with overlap by local regressor to process heterogeneous data, training a local regressor for each subset, and recording the regression result value of the first local regressor as ul(x) The gate network generates a gate function for each local regressor, and the gate function generated by the first gate network is recorded as pil(x) The formula of the age estimation result is:
where x is the input picture sample and y is the estimated age.
Specifically, the local regressor processes the heterogeneous data by dividing the training data into k subsets having overlap, such that each subset is used to train a local regressor, noting that the value of the l-th local regressor is ul(x) In order to effectively combine the results of the local regressors, the gate network generates a gate function for each local regressor, noting that the value of the first gate network is pil(x) The final age estimate is then:
(1) local regression device
The local regressors divide the entire data space into a plurality of subspaces, and each local regressor performs regression in one subspace. Thus, each local regressor is like a domain expert, each local regressor only has knowledge in a small regression space, different local regressors cover different regression intervals, and the knowledge of the local regressors is collected to provide a robust and accurate result.
The classification is carried out according to age labels, each local regressor is responsible for the regression of an age interval, and the regression length of all the local regressors is the same and is uniformly distributed on the whole regression interval. To further model the continuity of the age labels, a high degree of overlap between local regressors was allowed. That is, two adjacent local regressors have high overlap with the corresponding responsible regression regions, which makes the adjacent local regressors have high similarity. Thus for any one value, there are multiple local regressors responsible for regressing it, which allows further use of ensemble learning to obtain more accurate results.
(2) Door network
The local regressor generates a regression result ul(x) To better combine the regression results, the gate network generates a gate function pi for each local regressorl(x) To perform the weighting. The process of generating the gate function by the gate network is by means of a bridge tree structure.
The proposed bridge tree network structure is presented below. As shown in fig. 3 and 4, two examples of constructing a bridge tree are given, respectively. Fig. 3 shows a process of constructing a 4-level binary bridge tree from a 4-level binary tree, and fig. 4 shows a process of constructing a 3-level trifurcated bridge tree from a 3-level trifurcated tree. The construction of a bridge tree is a process of connecting the tree structure by applying bridges layer by layer. The bridge connection is that for two adjacent nodes on the same layer in the tree structure, the rightmost child node of the left node and the leftmost child node of the right node are merged into one node. By applying bridge connection, the obtained bridge tree can effectively model continuity between adjacent nodes. The bridge tree structure has high flexibility, and the number of layers and the number of branches can be set according to problems, so that different specific structures can be obtained.
Based on a bridged tree structure, this structure can explicitly model continuity, making the structure more suitable for use in regression problems than a tree structure.
The network comprises two parts, namely a gate network and a local regressor. The local regressors can better model heterogeneous data, and the gate network uses a bridge tree structure, so that the continuity relation among the local regressors can be better modeled, the continuity relation in age estimation can be effectively modeled by the final bridge neural network, and better performance is obtained.
Next, how the gate network generates the gate function pi by the bridge tree structure is explainedl(x) In that respect The bridge tree has two types of nodes, one internal (decision node) and one leaf (prediction node). A gate function is generated on each leaf node, so that the leaf nodes are in one-to-one correspondence with the local regressors. The generation of the gate function on the leaf node is a top-down step-by-step decision process. Starting from the root node of the bridge tree, for a given sample x, each time it encounters a decision node, it is sent to the child nodes of that node. Each decision node has B child nodes, and a decision is made on which node to reach next, and in the embodiment of the present invention, a soft decision method is adopted, that is, the decision is distributed to each edge to the child nodes according to probability, so that
Representing the probability value on the edge from node o to node m, that is, when node o is reached, to
Reaches node m, then naturally:
and
wherein C isoRepresenting the set of children of node o. Once a leaf node l is reached, the gate function pi on the leaf node ll(x) It can be obtained by summing the path probability values of all paths from the root node to the leaf node/.
Further, in one embodiment of the invention, a gate function π is generated in the gate network through a bridge tree structurel(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value on the edge from node m to node n.
Specifically, a recursive formula of the leaf node gate function is given by generalizing the concept of the gate function to all nodes:
where n is0Representing the root node, FnRepresenting the parent of node n. Thus, all the edges on the bridge tree structure are given
The value of the gate function on the leaf node can be calculated.
Further, in the embodiment of the present invention, the full-link layer is used to implement the local regressors, the sigmoid function is used as the activation function of the full-link layer, and then each local regressor maps its activation value to its regression interval, so as to obtain the regression value of each local regressor. During the training process, the loss function of this part can be defined as:
Lreg(x，y)＝∑lIl(x，y)(y-ul(x))2，
wherein Il(x, y) indicates whether the sample label y falls within the regression interval for which the ith regressor is responsible, if so, it is 1, otherwise it is 0.
As shown in fig. 5, showing implementation details of the gate network, the gate network is similarly implemented using a fully-connected layer, each neural node representing an edge on the bridge tree. The B neural nodes are normalized by using a softmax layer, so that the probability values of all edges on the bridge tree can be obtained
To train the gate network, since there is no real target value for the gate network, an approximate target value is constructed:
using R ═ ΣlI(x,y)R＝∑lII (x, y) is used to feedAnd (6) row normalization. Training of the gate network uses KL divergence as a loss function:
finally, an overall loss function is defined to jointly train the gate function and the local regressor:
Ltotal(x,y)＝Lreg(x,y)+λLgate(x,y)
λ is used to balance the two loss functions.
The embodiment of the invention adopts a human face age automatic estimation framework based on a bridge type neural network, the framework firstly aligns the human face by using a human face alignment technology, then extracts the characteristics by using a convolution neural network, and then sends the characteristics into the bridge type neural network to obtain the age estimation value, and the whole process can be trained end to end, so that the robust and accurate age estimation value is obtained.
According to the human face age estimation method based on the bridge neural network, provided by the embodiment of the invention, the continuity relation among local regressors is explicitly modeled by providing the bridge network for sensing continuity, so that the final age estimation performance is improved by fully utilizing the information, and the accuracy is improved.
Next, a facial age estimation apparatus based on a bridge neural network according to an embodiment of the present invention will be described with reference to the drawings.
Fig. 6 is a schematic structural diagram of a facial age estimation device based on a bridge neural network according to an embodiment of the present invention.
As shown in fig. 6, the face age estimation apparatus 10 includes: a first acquisition module 100, a second acquisition module 200, an alignment module 300, and an estimation module 400.
The first obtaining module 100 is configured to obtain an input map to be estimated.
The second obtaining module 200 is configured to obtain a face region and a plurality of key points of a face according to an input picture by using a face detection technology.
The alignment module 300 is configured to perform face alignment according to the plurality of key points to obtain an aligned face picture.
The estimation module 400 is configured to input the aligned face image into a deep convolutional neural network for feature extraction, connect the full connection layer for feature extraction to a local regressor and a gate network of a bridge network, generate a regression result and a gate function, and perform weighting on the regression result and the gate function to obtain an age estimation result.
The human face age estimation device 10 can solve the problem of low accuracy rate in the existing age estimation, and through providing a bridge network for sensing continuity, the continuity relation between local regressors is explicitly modeled, so that the final age estimation performance is improved by fully utilizing the information, and the accuracy rate is improved.
Further, in an embodiment of the present invention, the estimation module is further configured to:
dividing training data into k subsets with overlap by local regressor to process heterogeneous data, training a local regressor for each subset, and recording the regression result value of the first local regressor as ul(x) The gate network generates a gate function for each local regressor, and the gate function generated by the first gate network is recorded as pil(x) The formula of the age estimation result is:
where x is the input picture sample and y is the estimated age.
Further, in one embodiment of the present invention, the local regressors divide the entire data space into a plurality of subspaces, each local regressor performing a regression in one of the subspaces.
Further, in one embodiment of the present invention, the local regressor generates a regression result value of ul(x) And the gate network generates a gate function pi corresponding to each local regressorl(x) To perform weighting.
Further, in one embodiment of the invention, the gates are generated in the gate network by a bridge tree structureFunction pil(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value on the edge from node m to node n.
It should be noted that the above explanation of the embodiment of the method for estimating the face age is also applicable to the apparatus of the embodiment, and is not repeated herein.
According to the human face age estimation device based on the bridge neural network, provided by the embodiment of the invention, the continuity relation among local regressors is explicitly modeled by providing the bridge network for sensing continuity, so that the final age estimation performance is improved by fully utilizing the information, and the accuracy is improved.
"first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (4)
1. A face age estimation method based on a bridge neural network is characterized by comprising the following steps:
acquiring an input picture to be estimated;
acquiring a face area and a plurality of key points of a face by a face detection technology according to the input picture;
performing face alignment according to the plurality of key points to obtain an aligned face picture; and
inputting the aligned face picture into a deep convolutional neural network for feature extraction, connecting a full-connection layer for feature extraction to a local regressor and a gate network of a bridge network to generate a regression result and a gate function, and weighting the regression result and the gate function to obtain an age estimation result;
the bridge network structure comprises the local regressors and the gate networks, the local regressors and the gate networks are realized through a full connection layer, each local regressor generates a regression result, the gate networks generate a gate function corresponding to each local regressor, and the gate networks generate the gate functions according to the bridge tree structure of the bridge network, wherein the bridge tree structure comprises internal nodes and leaf nodes, and each leaf node generates a gate function;
generating the gate function pi in the gate network by a bridge tree structurel(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value from node m to the edge of node n;
weighting the regression result and the gate function to obtain an age estimation result, further comprising:
dividing training data into k subsets with overlap by the local regressor to process heterogeneous data, training one local regressor for each subset, and recording the regression result value of the ith local regressor as ul(x) Noting that the gate function generated by the ith gate network is pil(x) The formula of the age estimation result is as follows:
where x is the input picture sample and y is the estimated age.
2. The method of claim 1, wherein the local regressors divide the entire data space into a plurality of subspaces, and wherein each local regressor performs regression in one of the subspaces.
3. A human face age estimation device based on a bridge type neural network is characterized by comprising:
the first acquisition module is used for acquiring an input picture to be estimated;
the second acquisition module is used for acquiring a face area and a plurality of key points of a face according to the input picture by a face detection technology;
the alignment module is used for carrying out face alignment according to the plurality of key points so as to obtain an aligned face picture;
the estimation module is used for inputting the aligned face pictures into a deep convolutional neural network for feature extraction, connecting the full-connection layer for feature extraction to a local regressor and a gate network of a bridge network to generate a regression result and a gate function, and weighting and summing the regression result and the gate function to obtain an age estimation result;
the bridge network structure comprises the local regressors and the gate networks, the local regressors and the gate networks are realized through a full connection layer, each local regressor generates a regression result, the gate networks generate a gate function corresponding to each local regressor, and the gate networks generate the gate functions according to the bridge tree structure of the bridge network, wherein the bridge tree structure comprises internal nodes and leaf nodes, and each leaf node generates a gate function;
generating the gate function pi in the gate network by a bridge tree structurel(x) And then, a recursion calculation formula of the leaf node gate function is given by popularizing the gate function to all nodes:
wherein n is0A root node is represented as a root node,
is n th0Gate function values, F, generated by the individual gate networknRepresenting the parent of the node n,
representing the probability value from node m to the edge of node n;
weighting the regression result and the gate function to obtain an age estimation result, further comprising:
dividing training data into k subsets with overlap by the local regressor to process heterogeneous data, training one local regressor for each subset, and recording the regression result value of the ith local regressor as ul(x) Noting that the gate function generated by the ith gate network is pil(x) The formula of the age estimation result is as follows:
where x is the input picture sample and y is the estimated age.
4. The apparatus of claim 3, wherein the local regressors divide the entire data space into a plurality of subspaces, and wherein each local regressor performs regression in one of the subspaces.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910121339.XA CN109886183B (en) | 2019-02-19 | 2019-02-19 | Human face age estimation method and device based on bridge type neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910121339.XA CN109886183B (en) | 2019-02-19 | 2019-02-19 | Human face age estimation method and device based on bridge type neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886183A CN109886183A (en) | 2019-06-14 |
| CN109886183B true CN109886183B (en) | 2021-01-12 |
Family
ID=66928434
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910121339.XA Active CN109886183B (en) | 2019-02-19 | 2019-02-19 | Human face age estimation method and device based on bridge type neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886183B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7714736B2 (en) * | 2007-10-30 | 2010-05-11 | Gm Global Technology Operations, Inc. | Adaptive filter algorithm for estimating battery state-of-age |
| US8520906B1 (en) * | 2007-09-24 | 2013-08-27 | Videomining Corporation | Method and system for age estimation based on relative ages of pairwise facial images of people |
| CN104869096A (en) * | 2015-04-30 | 2015-08-26 | 南京信息职业技术学院 | Bootstrap-based BPSK signal blind processing result credibility test method |
| CN106068515A (en) * | 2014-03-06 | 2016-11-02 | 高通股份有限公司 | Multiple spectra ultrasonic imaging |
| CN106503623A (en) * | 2016-09-27 | 2017-03-15 | 中国科学院自动化研究所 | Facial image age estimation method based on convolutional neural networks |
| CN107045622A (en) * | 2016-12-30 | 2017-08-15 | 浙江大学 | The face age estimation method learnt based on adaptive age distribution |
| CN107895214A (en) * | 2017-12-08 | 2018-04-10 | 北京邮电大学 | A kind of multivariate time series Forecasting Methodology |
| CN108171209A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of face age estimation method that metric learning is carried out based on convolutional neural networks |
| CN108447074A (en) * | 2018-02-02 | 2018-08-24 | 中国科学院西安光学精密机械研究所 | Underwater target identification method based on bidirectional self-adaptive semantic fusion |
| CN108734331A (en) * | 2018-03-23 | 2018-11-02 | 武汉理工大学 | Short-term photovoltaic power generation power prediction method based on LSTM and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030073255A1 (en) * | 2001-10-12 | 2003-04-17 | Sundar Narayanan | Novel self monitoring process for ultra thin gate oxidation |
| US10235467B2 (en) * | 2015-12-11 | 2019-03-19 | Samsung Electronics Co., Ltd. | Providing search results based on an estimated age of a current user of a mobile computing device |
-
2019
- 2019-02-19 CN CN201910121339.XA patent/CN109886183B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8520906B1 (en) * | 2007-09-24 | 2013-08-27 | Videomining Corporation | Method and system for age estimation based on relative ages of pairwise facial images of people |
| US7714736B2 (en) * | 2007-10-30 | 2010-05-11 | Gm Global Technology Operations, Inc. | Adaptive filter algorithm for estimating battery state-of-age |
| CN106068515A (en) * | 2014-03-06 | 2016-11-02 | 高通股份有限公司 | Multiple spectra ultrasonic imaging |
| CN104869096A (en) * | 2015-04-30 | 2015-08-26 | 南京信息职业技术学院 | Bootstrap-based BPSK signal blind processing result credibility test method |
| CN106503623A (en) * | 2016-09-27 | 2017-03-15 | 中国科学院自动化研究所 | Facial image age estimation method based on convolutional neural networks |
| CN107045622A (en) * | 2016-12-30 | 2017-08-15 | 浙江大学 | The face age estimation method learnt based on adaptive age distribution |
| CN107895214A (en) * | 2017-12-08 | 2018-04-10 | 北京邮电大学 | A kind of multivariate time series Forecasting Methodology |
| CN108171209A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of face age estimation method that metric learning is carried out based on convolutional neural networks |
| CN108447074A (en) * | 2018-02-02 | 2018-08-24 | 中国科学院西安光学精密机械研究所 | Underwater target identification method based on bidirectional self-adaptive semantic fusion |
| CN108734331A (en) * | 2018-03-23 | 2018-11-02 | 武汉理工大学 | Short-term photovoltaic power generation power prediction method based on LSTM and system |
Non-Patent Citations (3)
| Title |
|---|
| Accuracy and sampling error of two age estimation techniques using rib histomorphometry on a modern sample;Julieta G. García-Donas等;《Journal of Forensic and Legal Medicine》;20151025;第28-34页 * |
| 中国人渗出性年龄相关性黄斑变性易感基因及与环境交互作用研究;楚洁;《中国博士学位论文全文数据库（医药卫生科技辑）》;20091215;第E073-21页 * |
| 计算机应用;郑毅等;《计算机应用》;20180610;第38卷(第6期);第1568-1574页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886183A (en) | 2019-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914644B (en) | Dual-mode cooperation based weak supervision time sequence action positioning method and system | |
| Xia et al. | W-net: A deep model for fully unsupervised image segmentation | |
| CN109002845A (en) | Fine granularity image classification method based on depth convolutional neural networks | |
| Yan et al. | Chiller faults detection and diagnosis with sensor network and adaptive 1D CNN | |
| CN114220271A (en) | Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network | |
| US20190303713A1 (en) | Discovery of shifting patterns in sequence classification | |
| CN111090764A (en) | Image classification method and device based on multitask learning and graph convolution neural network | |
| CN110826702A (en) | Abnormal event detection method for multitask deep network | |
| CN112580902B (en) | Object data processing method and device, computer equipment and storage medium | |
| CN112836105B (en) | Large-scale student aerobic capacity clustering method based on movement physiological characterization fusion | |
| CN113283282B (en) | Weak supervision time sequence action detection method based on time domain semantic features | |
| CN111047078A (en) | Traffic characteristic prediction method, system and storage medium | |
| CN112668438A (en) | Infrared video time sequence behavior positioning method, device, equipment and storage medium | |
| CN113947157A (en) | Dynamic brain effect connection network generation method based on hierarchical clustering and structural equation model | |
| Wang et al. | Correlated time series self-supervised representation learning via spatiotemporal bootstrapping | |
| Gorodnichev et al. | Research and Development of a System for Determining Abnormal Human Behavior by Video Image Based on Deepstream Technology | |
| CN116052254A (en) | Visual continuous emotion recognition method based on extended Kalman filtering neural network | |
| CN111586728A (en) | Small sample characteristic-oriented heterogeneous wireless network fault detection and diagnosis method | |
| Anıl et al. | Deep learning based prediction model for the next purchase | |
| CN109886183B (en) | Human face age estimation method and device based on bridge type neural network | |
| CN115375038A (en) | Aircraft engine failure mode identification and service life prediction method | |
| Awad et al. | Prediction of general high school exam result level using multilayer perceptron neural networks | |
| KR102340387B1 (en) | Method of learning brain connectivity and system threrfor | |
| Gerych et al. | Positive unlabeled learning with a sequential selection bias | |
| CN112185543A (en) | Construction method of medical induction data flow classification model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |