CN112215076A - Deep handwriting identification method and device based on double-tower network - Google Patents
Deep handwriting identification method and device based on double-tower network Download PDFInfo
- Publication number
- CN112215076A CN112215076A CN202010954616.8A CN202010954616A CN112215076A CN 112215076 A CN112215076 A CN 112215076A CN 202010954616 A CN202010954616 A CN 202010954616A CN 112215076 A CN112215076 A CN 112215076A
- Authority
- CN
- China
- Prior art keywords
- handwriting
- network
- channel
- data
- image
- 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.)
- Granted
Links
Images
Classifications
-
- 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
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/33—Writer recognition; Reading and verifying signatures based only on signature image, e.g. static signature recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The embodiment of the invention provides a method and a device for deep handwriting authentication based on a double-tower network. In the identification process, noise interference in the first handwriting image and the second handwriting image is removed in a data reconstruction mode, robustness of the first handwriting image and the second handwriting image is improved, due to weight sharing in a trained double-tower network model, characteristics of global handwriting data are paid more attention to, similar data are closer, differences among dissimilar data are increased, generalization capability of the first handwriting image and the second handwriting image is improved, after data reconstruction is carried out, the first characteristic data and the second characteristic data of the reconstructed data are fused, characteristic data values of the similar images are increased, and identification results are more accurate.
Description
Technical Field
The invention belongs to the field of deep neural network identification, and particularly relates to a deep handwriting identification method and device based on a double-tower network.
Background
The adult character writing process is trained for a long time, and a stable writing style is formed. The written characters have biological behavior characteristics, and play an increasingly important role in the fields of judicial assessment, medical disputes and the like. In these fields, it is often necessary to authenticate written text, a process known as handwriting authentication.
The prior art handwriting authentication process is as follows:
the method comprises the steps of collecting a plurality of characters written by people, converting the characters into pictures, obtaining a plurality of sample sets containing picture data, inputting the sample sets into a deep learning network model, and training the deep learning network model in an iterative mode until a training cutoff condition is reached. And using the trained deep learning network model to identify whether the character to be determined is written by the corresponding person, thereby completing handwriting identification.
Due to the complex background of the text to be authenticated, the text in the loan has a large influence on the signature, assuming that the text to be authenticated is a signature appearing in a manually written loan. If the signature is a signature highly simulating a corresponding person, the identification accuracy is greatly influenced, the similarity degree between similar handwriting is emphatically trained when the neural network model is trained so as to achieve the accurate identification effect, so that the generalization capability and the robustness of the trained neural network model are poor, namely certain interference is caused on the handwriting not contained in the sample set and the interference in the handwriting, and the identification accuracy of the neural network model is reduced.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method and a device for deep handwriting authentication based on a double-tower network, which solve the problem of low accuracy of handwriting authentication. The technical problem to be solved by the invention is realized by the following technical scheme:
in a first aspect, an embodiment of the present invention provides a deep handwriting authentication method based on a double-tower network, including:
acquiring a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated;
inputting the first handwriting image into a first input channel of the trained double-tower network model and inputting a second handwriting image into a second input channel of the trained double-tower network model to obtain a recognition result output by the trained double-tower network model; the recognition result comprises: probability of a handwriting belonging to a person to be authenticated;
determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated;
the trained double-tower network model is obtained by training a preset double-tower network model, and is used for carrying out data reconstruction on the first handwriting image and the second handwriting image, extracting first characteristic data from data of the reconstructed first handwriting image, extracting second characteristic data from data of the reconstructed second handwriting image, fusing the first characteristic data and the second characteristic data, and outputting a recognition result based on the fused characteristic data.
Optionally, before the first handwriting image is input into the first input channel of the trained double-tower network model and the second handwriting image is input into the second input channel of the trained double-tower network model, the deep handwriting authentication method provided in the embodiment of the first aspect of the present invention further includes:
respectively binarizing the first handwriting image and the second handwriting image;
performing image channel conversion on the binarized first handwriting image and the binarized second handwriting image so as to enable an image channel of the first handwriting image to be matched with an image channel required by the first input channel or an image channel of the second handwriting image to be matched with an image channel required by the second input channel;
wherein the image channel required by the first input channel is the same as the image channel required by the second input channel.
Optionally, the preset double-tower network model includes: the system comprises an encoding and decoding network, a backbone network, a feature crossing network and a classification output network, wherein an input channel of the encoding and decoding network comprises: the encoding and decoding network is used for reconstructing data of an input sample image, the backbone network extracts feature data after image information is reconstructed, the feature crossing network is used for performing feature difference calculation on the feature data of the two channel networks of the backbone network, and an identification result is output based on a calculation result of the feature difference;
the characteristic data includes: the first characteristic data and the second characteristic data.
Optionally, the encoding and decoding network includes: a first channel network and a second channel network, the first channel network and the second channel network sharing a weight, the backbone network comprising: the system comprises a third channel network and a fourth channel network, wherein the third channel network and the fourth channel network share weight, the output of the first channel network is connected with the input of the third channel network, the output of the second channel network is connected with the fourth channel network, and the output of the third channel network is connected with the output of the fourth channel network and the input connected with the characteristic cross network.
Optionally, the step of training the preset double-tower network model includes:
the method comprises the following steps: constructing a sample set, the sample set comprising: a positive sample pair and a negative sample pair, the positive sample pair and the negative sample pair being a plurality of samples selected from a corresponding data set;
step two: for a sample pair in the sample set, inputting a first sample in the sample pair into the first channel network and inputting a second sample in the sample pair into the second channel network, and obtaining an identification result of the feature crossing network;
the first channel network is configured to transmit a first result of summation of reconstructed data and data of the first sample to the third channel network after data reconstruction is performed on the first sample, the second channel network is configured to transmit a second result of summation of reconstructed data and data of the second sample to the fourth channel network after data reconstruction is performed on the second sample, the third channel network is configured to extract first feature data from the first result and transmit the first feature data to the feature crossing network, the fourth channel network is configured to extract second feature data from the second result and transmit the second feature data to the feature crossing network, the feature crossing network is configured to fuse the first feature data and the second feature data, and an identification result is output based on the fused result;
step three: calculating a contrast loss based on the first result and the second result;
step four: calculating the cross entropy loss of the preset double-tower network model based on the recognition result;
step five: adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss, and repeating the second step to the fourth step until a training cut-off condition is reached;
step six: and determining the preset double-tower network model reaching the training cut-off condition as a trained double-tower network model.
Optionally, the step of constructing a sample set includes:
acquiring a first data set of a plurality of handwriting images of the same person and a second data set of a plurality of handwriting images of different persons;
randomly selecting a preset number of positive sample pairs from the first data set;
randomly selecting a preset number of negative sample pairs from the second data set;
forming a set of samples using the pair of positive samples and the pair of negative samples.
Optionally, the step of adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss includes:
weighting and summing the contrast loss and the cross entropy loss to obtain a total loss value;
and adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value.
Optionally, the step of adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value includes:
and adjusting the weights of the coding and decoding network, the backbone network and the feature crossing network in the preset double-tower network model by using a back propagation algorithm according to the direction of reducing the total loss value.
In a second aspect, an embodiment of the present invention provides a deep handwriting authentication apparatus based on a double-tower network, including:
the system comprises an acquisition module, a verification module and a verification module, wherein the acquisition module is used for acquiring a first handwriting image to be verified and a second handwriting image of a person to be verified;
the recognition module is used for inputting the first handwriting image into a first input channel of the trained double-tower network model and inputting a second handwriting image into a second input channel of the trained double-tower network model, and recognizing the first handwriting image and the second handwriting image by using the trained double-tower network model to obtain a recognition result; the recognition result is as follows: probability of whether the person belongs to the handwriting of the person to be authenticated;
the determining module is used for determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated;
the trained double-tower network model is obtained by training a preset double-tower network model, and is used for carrying out data reconstruction on the first handwriting image and the second handwriting image, extracting first characteristic data from data of the reconstructed first handwriting image, extracting second characteristic data from data of the reconstructed second handwriting image, fusing the first characteristic data and the second characteristic data, and outputting a recognition result based on the fused characteristic data.
Optionally, the preset double-tower network model includes: the system comprises an encoding and decoding network, a backbone network, a feature crossing network and a classification output network, wherein an input channel of the encoding and decoding network comprises: the encoding and decoding network is used for reconstructing data of an input sample image, the backbone network extracts feature data after image information is reconstructed, and the feature crossing network is used for performing feature difference calculation on the feature data of the two channel networks of the backbone network and outputting an identification result based on a calculation result of the feature difference.
Optionally, the device for deep handwriting authentication based on the double-tower network provided in the ninth embodiment of the present invention further includes:
the drying module is used for respectively carrying out binarization on the first handwriting image and the second handwriting image;
performing image channel conversion on the binarized first handwriting image and the binarized second handwriting image so as to enable an image channel of the first handwriting image to be matched with an image channel required by the first input channel or an image channel of the second handwriting image to be matched with an image channel required by the second input channel;
wherein the image channel required by the first input channel is the same as the image channel required by the second input channel.
Optionally, the encoding and decoding network includes: a first channel network and a second channel network, the first channel network and the second channel network sharing a weight, the backbone network comprising: the system comprises a third channel network and a fourth channel network, wherein the third channel network and the fourth channel network share weight, the output of the first channel network is connected with the input of the third channel network, the output of the second channel network is connected with the fourth channel network, and the output of the third channel network is connected with the output of the fourth channel network and the input connected with the characteristic cross network.
Optionally, the step of training the preset double-tower network model includes:
the method comprises the following steps: constructing a sample set, the sample set comprising: a positive sample pair and a negative sample pair, the positive sample pair and the negative sample pair being a plurality of samples selected from a corresponding data set;
step two: for each positive sample pair and each negative sample pair, inputting the positive sample pair into the first channel network and inputting the negative sample pair into the second channel network, and obtaining the identification result of the feature crossing network;
the first channel network is configured to transmit a first result of summation of reconstructed data and data of the first sample to the third channel network after data reconstruction is performed on the first sample, the second channel network is configured to transmit a second result of summation of reconstructed data and data of the second sample to the fourth channel network after data reconstruction is performed on the second sample, the third channel network is configured to extract first feature data from the first result and transmit the first feature data to the feature crossing network, the fourth channel network is configured to extract second feature data from the second result and transmit the second feature data to the feature crossing network, the feature crossing network is configured to fuse the first feature data and the second feature data, and an identification result is output based on the fused result;
step three: calculating a contrast loss based on the first result and a cross-entropy loss based on the second result;
step four: calculating a third loss value of the preset double-tower network model based on the recognition result;
step five: adjusting the weight in the preset double-tower network model based on the contrast loss, the cross entropy loss and the third loss value, and repeating the second step to the fourth step until a training cut-off condition is reached;
step six: and determining the preset double-tower network model reaching the training cut-off condition as a trained double-tower network model.
Optionally, the step of constructing a sample set includes:
acquiring a first data set of a plurality of handwriting images of the same person and a second data set of a plurality of handwriting images of different persons;
randomly selecting a preset number of positive sample pairs from the first data set;
randomly selecting a preset number of negative sample pairs from the second data set;
forming a set of samples using the pair of positive samples and the pair of negative samples.
Optionally, the step of adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss includes:
weighting and summing the contrast loss and the cross entropy loss to obtain a total loss value;
and adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value.
Optionally, the step of adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value includes:
and adjusting the weights of the coding and decoding network, the backbone network and the feature crossing network in the preset double-tower network model by using a back propagation algorithm according to the direction of reducing the total loss value.
Optionally, the determining, according to the probability of the handwriting belonging to the person to be authenticated, whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated includes:
and when the probability of the handwriting belonging to the person to be authenticated in the recognition result is greater than a preset threshold value, determining that the handwriting in the first handwriting image is the handwriting of the person to be authenticated.
After a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated are obtained, the first handwriting image and the second handwriting image are input into a trained double-tower network model, and an identification result output by the double-tower network model is obtained. The trained double-tower network model removes noise interference in the first handwriting image and the second handwriting image in a data reconstruction mode in the recognition process, robustness of the trained double-tower network model is improved, the trained double-tower network model focuses more on the features of the global handwriting data, similar data are closer, difference between dissimilar data is increased, generalization capability of the trained double-tower network model is improved, and further after data reconstruction is carried out, the first feature data and the second feature data extracted from the reconstructed data are fused, so that feature data values of the similar images are increased, and recognition results are more accurate.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
FIG. 1 is a flowchart of a deep handwriting authentication method based on a double-tower network according to an embodiment of the present invention;
fig. 2 is an effect diagram of converting a binarized image into a three-channel image according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a double tower network model provided by an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a feature crossover network provided by an embodiment of the present invention;
fig. 5 is a structural diagram of a deep handwriting authentication device based on a double-tower network according to an embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
As shown in fig. 1, a method for deep handwriting authentication based on a double-tower network provided by an embodiment of the present invention includes:
s11, acquiring a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated;
it is understood that the person to be authenticated refers to a person who writes the handwriting to be authenticated, and the person who writes the handwriting to be authenticated refers to a person who may write the handwriting to be authenticated. The handwriting to be authenticated refers to the handwriting formed by characters, characters and figures which need to be authenticated. The first handwriting image can be obtained by converting the handwriting to be authenticated into a picture, and the second handwriting image can be obtained by converting the handwriting written by the person to be authenticated into the picture. Of course, in the conversion process, photographing, scanning and the like can be used, and the invention is not limited herein.
For example, a person signature exists in the borrowing data, the person denies that the person who has signed the borrowing data, the person who is the person to be authenticated, the person signature on the borrowing data is handwriting to be authenticated, the handwriting to be authenticated is converted into a first handwriting image, and the signature written by the person himself is used as a second handwriting image.
S12, inputting the first handwriting image into a first input channel of the trained double-tower network model and inputting the second handwriting image into a second input channel of the trained double-tower network model, and recognizing the first handwriting image and the second handwriting image by using the trained double-tower network model to obtain a recognition result;
wherein, the recognition result includes: probability of a handwriting belonging to a person to be authenticated; the trained double-tower network model is obtained by training a preset double-tower network model, the trained double-tower network model is used for carrying out data reconstruction on the first handwriting image and the second handwriting image, during data reconstruction, the weight of the trained double-tower network model is shared, first characteristic data is extracted from the data of the reconstructed first handwriting image, second characteristic data is extracted from the data of the reconstructed second handwriting image, the first characteristic data and the second characteristic data are fused, and a recognition result is output based on the fused characteristic data.
It can be understood that, in the process of reconstructing data by the double-tower network model, the first handwriting image and the second handwriting image need to be encoded and decoded, and the coding and decoding will remove the noise interference in the first handwriting image and the second handwriting image, so the robustness of the double tower network model can be enhanced, meanwhile, the weight is shared when the data of the two-way input channel is reconstructed, so that the double-tower network model can focus on the characteristics of the global handwriting data, the similar data are closer, the difference between the dissimilar data is increased, thus, if the first handwriting image is similar to the second handwriting image, the closer the values of the data of the two reconstructed images are, if the first handwriting image is not similar to the second handwriting image, the difference of the values of the data of the two images is larger, so that the generalization capability of the double-tower network model is enhanced. And then, after data reconstruction is carried out, fusing the first characteristic data and the second characteristic data of the data after reconstruction is extracted, so that the characteristic data value of the similar image is increased, the accuracy of the recognition result is increased based on the fused characteristic data, and the accuracy of the recognition result is output.
And S13, determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the recognition result.
It can be understood that the higher the probability of the handwriting belonging to the same person in the recognition result is, the higher the similarity between the first handwriting image and the second handwriting image is, the higher the probability that the handwriting to be authenticated is written by the person to be authenticated is, and whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated can be determined through the recognition result.
After a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated are obtained, the first handwriting image and the second handwriting image are input into a trained double-tower network model, and an identification result output by the double-tower network model is obtained. In the recognition process of the trained double-tower network model, noise interference in the first handwriting image and the second handwriting image is removed in a data reconstruction mode, robustness of the double-tower network model is improved, weight sharing is performed in the trained double-tower network model, characteristics of global handwriting data are paid more attention to, similar data are closer, difference between dissimilar data is increased, generalization capability of the double-tower network model is improved, further, after data reconstruction is performed, the first characteristic data and the second characteristic data of the reconstructed data are extracted and fused, characteristic data values of the similar images are increased, and recognition results are more accurate.
Example two
As an optional implementation manner provided by the embodiment of the present invention, before the step S12, the method for deep-handwriting authentication based on a double-tower network according to the embodiment of the present invention further includes:
the method comprises the following steps: respectively binarizing the first handwriting image and the second handwriting image;
step two: performing image channel conversion on the binarized first handwriting image and the binarized second handwriting image so as to enable an image channel of the first handwriting image to be matched with an image channel required by a first input channel or an image channel of the second handwriting image to be matched with an image channel required by a second input channel;
the image channel required by the first input channel is the same as the image channel required by the second input channel.
It can be understood that when a handwriting picture is obtained, the background color of the obtained picture is often greatly different due to the difference of the background colors of the paper, and the difference is often generated when the handwriting of the same person is collected, so that the difference of the background colors of the handwriting pictures of different persons can be learned by the double-tower network model instead of the difference of the real handwriting characteristics, and the background of the handwriting picture needs to be removed.
Firstly, after a first handwriting image and a second handwriting image are obtained, graying operation is carried out on the first handwriting image and the second handwriting image, then binarization operation is carried out on the grayed first handwriting image and the second handwriting image by using a self-adaptive threshold value adaptive threshold algorithm, after the binarization operation, an image channel of the obtained image is a single channel, and an image needing to be input by a double-tower network model is not the single channel, so that the image needs to be converted back to the number of the image channels required by the double-tower network model after binarization. For example, if a three-channel image is required for the double-tower network model, the binarized image is converted into a three-channel image, and the conversion effect is shown in fig. 2.
According to the method and the device, the background removing processing is carried out on the first note image and the second note image, so that the interference of a background part on the double-tower network model is reduced, and the accuracy of the recognition result can be improved.
EXAMPLE III
As an optional implementation manner of the embodiment of the present invention, the preset double-tower network model includes: the system comprises an encoding and decoding network, a backbone network, a feature crossing network and a classification output network, wherein an input channel of the encoding and decoding network comprises: the system comprises a first input channel and a second input channel, wherein the coding and decoding network is used for carrying out data reconstruction on an input sample image, the backbone network extracts characteristic data after image information is reconstructed, the characteristic cross network is used for carrying out characteristic difference calculation on the characteristic data of two channel networks of the backbone network, and an identification result is output based on a calculation result of the characteristic difference;
wherein the characteristic data includes: first characteristic data and second characteristic data. The encoding and decoding network includes: a first channel network and a second channel network, the first channel network and the second channel network sharing a weight, the backbone network comprising: the third channel network and the fourth channel network share the weight, the output of the first channel network is connected with the input of the third channel network, the output of the second channel network is connected with the fourth channel network, and the output of the third channel network is connected with the output of the fourth channel network and the input of the characteristic crossover network.
The following describes the structure of the double tower network model provided by the embodiment of the present invention by taking fig. 3 as an example. As shown in fig. 3, in this structure, picture 1 may be a first handwriting image, and then picture 2 is a second handwriting image, and picture 1 is input through a first channel network, which is composed of an encoding sub-network, a decoding sub-network and a summing sub-network connected to picture 1; the second channel network is composed of an encoding sub-network, a decoding sub-network and a summing sub-network connected with the picture 2; in fig. 3, the sub-network is a neuron network formed by connecting neurons with each other, and the summing sub-network is a network structure located between the decoding sub-network and the backbone network in fig. 3. In fig. 3, the first channel network and the second channel network have the same structure, and the first channel network and the second channel network share the weight therebetween. And connecting the output of the coding and decoding network with a backbone network backbone, sharing weight between a third channel network and a fourth channel network in the backbone network, converging the subsequent output in a feature cross network, performing feature cross fusion by the feature cross network after extracting feature data in the third channel network and the fourth channel network, and identifying whether the handwriting in the picture 1 and the handwriting in the picture 2 are written by the same person or not based on a feature fusion result.
It can be understood that the front part of the double-tower network model adopts an encoder-decoder structure, which is intended to reconstruct the features of the handwriting picture, and the (encoder-decoder) structure will forcedly remove the noise in the handwriting picture to obtain the input without noise, so that the method has better generalization and robustness compared with the method of directly using a backbone (neural network model) to perform feature extraction.
The encoder coding part and the decoder decoding part both use the rennet 18 to extract features, and the encoder coding of the two-way input shares weight, so that the encoder coding can better focus on all handwriting pictures, and the decoder decoding part also shares weight, so that the decoder can better focus on the features of the global handwriting data.
Example four
As an optional implementation manner provided by the embodiment of the present invention, the step of training the preset double-tower network model includes:
the method comprises the following steps: constructing a sample set, wherein the sample set comprises: a positive sample pair and a negative sample pair, the positive sample pair and the negative sample pair being a plurality of samples selected from a corresponding dataset;
it can be understood that the positive sample pair and the negative sample pair are constructed, and are simultaneously input into the double-tower network model to meet the input requirement of the double-tower network model, so that the double-tower network model does not depend on a single sample during learning, and the accuracy is improved.
Step two: aiming at a sample pair in a sample set, inputting a first sample in the sample pair into a first channel network and inputting a second sample in the sample pair into a second channel network to obtain a recognition result of a feature crossing network;
the first channel network is used for transmitting a first result of data summation of reconstructed data and the first sample to a third channel network after data reconstruction of the first sample is performed, the second channel network is used for transmitting a second result of data summation of the reconstructed data and the second sample to a fourth channel network after data reconstruction of the second sample is performed, the third channel network is used for extracting first feature data from the first result and transmitting the first feature data to a feature crossing network, the fourth channel network is used for extracting second feature data from the second result and transmitting the second feature data to the feature crossing network, the feature crossing network is used for fusing the first feature data and the second feature data, and an identification result is output based on the fused result;
the first characteristic data is data representing characteristic points of the first handwriting image, and the second characteristic data is data representing characteristic points of the second handwriting image. The dimensions of the first feature data and the second feature data are the same, the feature points represent features of the image, and the feature points can be corner points, edge points, SIFT feature points and the like. The invention is not limited thereto.
The structure of the feature crossbar network is shown in fig. 4, in which fea for the first feature data in fig. 41Indicating, fea for second characteristic data2Indicating that the feature crossing network is to be (fea)1-fea2)2And an
Indicating multiplication.
Exemplarily, assume fea1＝[1，2，2，4]，fea2＝[0，5，7，9]Then (fea)1-fea2)2＝[(1-0)2，(2-5)2，(2-7)2，(4-9)2]，
Step three: calculating a contrast loss based on the first result and the second result;
wherein, the calculation formula of the contrast loss is as follows:
L(x1,x2y) denotes the loss of contrast, x1Representing a feature in the first result, x2Representing features in the second sample result, D (x)1,x2)＝||x1-x2||2Representing two sample features x1And x2The euclidean distance of (a) y is a label indicating whether two samples are matched, y equals 1 and represents that the two samples are similar or matched, y equals 0 and represents mismatch, m is a set similarity threshold, N is the number of samples, and N represents the serial number of the samples.
When calculating the first contrast loss, because the decoding network (decoder) has a plurality of upsampling layers, the characteristics output by the upsampling layers are calculated in the form of characteristic pairs, so that a plurality of contrast losses are obtained, and the contrast losses of the upsampling layers are added to be used as the total contrast loss of the decoding network. Therefore, the characteristics which are possibly learned in each layer of the decoding network have distinguishing force (the distance between one person's handwriting is small, the distance between different persons' handwriting is large), the output of the decoding network (decoder) is consistent with the size of the image of the input coding network (encoder) after a plurality of times of upsampling, and then the characteristics output by the coding-decoding network (encoder-decoder) are added with the original image, so that the characteristics of the backbone network (backbone) to be input have stronger representation capability.
Step four: calculating the cross entropy loss of a preset double-tower network model based on the recognition result;
wherein, the calculation formula of the cross entropy loss is as follows:
l (y, p) represents cross entropy loss, y is a real label of the sample, p is the probability of the output of the double-tower network model, N is the number of the samples, and i represents the serial number of the samples.
Step five: adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss, and repeating the second step to the fourth step until a training cut-off condition is reached;
wherein the training cutoff conditions include: the number of iterations or penalty values weighted for contrast penalty, cross entropy penalty is minimized.
It can be understood that when the weight of the double-tower network model is adjusted, besides the cross entropy Loss based on the recognition result, the first contrast Loss of the encoding and decoding network output is considered at the same time, because the decoder decoding part calculates the contrast Loss between two paths of characteristics sampled on each layer of the decoder once, the difference of the handwriting data characteristics of the same person can be reduced, and the difference of the handwriting data characteristics between different persons is increased.
Step six: and determining the preset double-tower network model reaching the training cut-off condition as the trained double-tower network model.
EXAMPLE five
As an optional implementation manner of the embodiment of the present invention, the step of constructing the sample set includes:
the method comprises the following steps: acquiring a first data set of a plurality of handwriting images of the same person and a second data set of a plurality of handwriting images of different persons;
wherein the first data set comprises: and the data of the handwriting image of the same person comprises in the second data set: and the data of handwriting images of different people is three-channel RGB data of the original image.
Step two: randomly selecting a preset number of positive sample pairs from the first data set;
step three: randomly selecting a preset number of negative sample pairs from a second data set;
step four: the sample set is constructed using positive and negative sample pairs.
For example, 200 pieces of image data can be randomly taken from handwriting pictures of the same person as positive sample pairs, 200 pieces of image data can be randomly taken from handwriting pictures of different persons, and then the selected image data can be taken as negative sample pairs to construct a required sample set.
It can be understood that, before the first step of the fifth embodiment, the images of the positive sample pair and the images of the negative sample pair need to be binarized respectively; and carrying out image channel conversion on the binarized positive sample pair image and the binarized negative sample pair image so as to enable the image channel of the positive sample pair image to be matched with the image channel required by the first input channel or the image channel of the negative sample pair image to be matched with the image channel required by the second input channel.
EXAMPLE six
As an optional implementation manner of the embodiment of the present invention, the third step in the fourth embodiment of the present invention includes:
the method comprises the following steps: weighting and summing the contrast loss and the cross entropy loss to obtain a total loss value;
step two: and adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value.
EXAMPLE seven
As an optional implementation manner of the embodiment of the present invention, the second step in the sixth embodiment includes:
and adjusting the weights of the coding and decoding network, the backbone network, the characteristic cross network and the characteristic cross network in the preset double-tower network model by using a back propagation algorithm according to the direction of reducing the total loss value.
Example eight
As an optional implementation manner of the embodiment of the present invention, the step of S13 includes: and when the probability of the handwriting belonging to the same person in the recognition result is greater than a preset threshold value, determining that the handwriting in the first handwriting image is the handwriting of the person to be authenticated.
The preset threshold is a preset value, the value is not more than 1 and not less than 0.5, and is specifically set according to industry experience, and the preset threshold is 0.9, 0.85 and 0.95 in actual operation.
It can be understood that when the probability that the handwriting belongs to the same person in the recognition result is high, it indicates that the high probability of the handwriting in the handwriting image to be authenticated is written by the person to be determined, and when the probability that the handwriting belongs to the same person in the recognition result is low, it indicates that the high probability of the handwriting in the handwriting image to be authenticated is not written by the person to be determined. Therefore, according to the threshold comparison, it can be determined whether the handwriting image to be authenticated is the self handwriting of the person to be authenticated.
Example nine
As shown in fig. 5, an embodiment of the present invention provides a deep handwriting authentication apparatus based on a double-tower network, including:
an obtaining module 51, configured to obtain a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated;
the recognition module 52 is configured to input the first handwriting image into a first input channel of the trained double-tower network model and input the second handwriting image into a second input channel of the trained double-tower network model, and recognize the first handwriting image and the second handwriting image by using the trained double-tower network model to obtain a recognition result; the recognition result is: probability of whether the person belongs to the handwriting of the person to be authenticated;
the method comprises the steps that a trained double-tower network model is obtained by training a preset double-tower network model, the trained double-tower network model is used for carrying out data reconstruction on a first handwriting image and a second handwriting image, extracting first characteristic data from data of the reconstructed first handwriting image, extracting second characteristic data from data of the reconstructed second handwriting image, fusing the first characteristic data and the second characteristic data, and outputting a recognition result based on the fused characteristic data;
and the determining module 53 is used for determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated.
Example ten
Optionally, the preset double-tower network model includes: the system comprises an encoding and decoding network, a backbone network, a feature crossing network and a classification output network, wherein an input channel of the encoding and decoding network comprises: the system comprises a first input channel and a second input channel, wherein the coding and decoding network is used for carrying out data reconstruction on an input sample image, the backbone network extracts characteristic data after image information is reconstructed, and the characteristic cross network is used for carrying out characteristic difference calculation on the characteristic data of two channel networks of the backbone network and outputting an identification result based on a calculation result of the characteristic difference.
Optionally, the device for deep handwriting authentication based on the double-tower network provided in the ninth embodiment of the present invention further includes: a de-drying module for
Respectively binarizing the first handwriting image and the second handwriting image;
performing image channel conversion on the binarized first handwriting image and the binarized second handwriting image so as to enable an image channel of the first handwriting image to be matched with an image channel required by a first input channel or an image channel of the second handwriting image to be matched with an image channel required by a second input channel;
the image channel required by the first input channel is the same as the image channel required by the second input channel.
Optionally, the encoding and decoding network includes: a first channel network and a second channel network, the first channel network and the second channel network sharing a weight, the backbone network comprising: the system comprises a third channel network and a fourth channel network, wherein the third channel network and the fourth channel network share weight, the output of the first channel network is connected with the input of the third channel network, the output of the second channel network is connected with the fourth channel network, and the output of the third channel network is connected with the input of the fourth channel network, which is connected with the characteristic cross network.
Optionally, the step of training the preset double-tower network model includes:
the method comprises the following steps: constructing a sample set, wherein the sample set comprises: a positive sample pair and a negative sample pair, the positive sample pair and the negative sample pair being a plurality of samples selected from a corresponding dataset;
step two: aiming at a sample pair in a sample set, inputting a first sample in the sample pair into a first channel network and inputting a second sample in the sample pair into a second channel network to obtain a recognition result of a feature crossing network;
the first channel network is used for transmitting a first result of data summation of reconstructed data and the first sample to a third channel network after data reconstruction of the first sample is performed, the second channel network is used for transmitting a second result of data summation of the reconstructed data and the second sample to a fourth channel network after data reconstruction of the second sample is performed, the third channel network is used for extracting first feature data from the first result and transmitting the first feature data to a feature crossing network, the fourth channel network is used for extracting second feature data from the second result and transmitting the second feature data to the feature crossing network, the feature crossing network is used for fusing the first feature data and the second feature data, and an identification result is output based on the fused result;
step three: calculating a contrast loss based on the first result and the second result;
step four: calculating the cross entropy loss of a preset double-tower network model based on the recognition result;
step five: adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss, and repeating the second step to the fourth step until a training cut-off condition is reached;
step six: and determining the preset double-tower network model reaching the training cut-off condition as the trained double-tower network model.
Optionally, the step of constructing a sample set includes:
acquiring a first data set of a plurality of handwriting images of the same person and a second data set of a plurality of handwriting images of different persons;
randomly selecting a preset number of positive sample pairs from the first data set;
randomly selecting a preset number of negative sample pairs from a second data set;
the sample set is constructed using positive and negative sample pairs.
Optionally, the step of adjusting the weight in the preset double-tower network model based on the contrast loss, the cross entropy loss, and the third loss value includes:
weighting and summing the contrast loss and the cross entropy loss to obtain a total loss value;
and adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value.
Optionally, the step of adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value includes:
and adjusting the weights of the coding and decoding network, the backbone network and the characteristic cross network in the preset double-tower network model by using a back propagation algorithm according to the direction of reducing the total loss value.
Optionally, determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated comprises:
and when the probability of the handwriting belonging to the person to be authenticated in the recognition result is greater than a preset threshold value, determining that the handwriting in the first handwriting image is the handwriting of the person to be authenticated.
After a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated are obtained, the first handwriting image and the second handwriting image are input into a trained double-tower network model, and an identification result output by the double-tower network model is obtained. The trained double-tower network model removes noise interference in the first handwriting image and the second handwriting image in a data reconstruction mode in the recognition process, robustness of the trained double-tower network model is improved, the trained double-tower network model focuses more on the features of the global handwriting data, similar data are closer, difference between dissimilar data is increased, generalization capability of the trained double-tower network model is improved, and further after data reconstruction is carried out, the first feature data and the second feature data extracted from the reconstructed data are fused, so that feature data values of the similar images are increased, and recognition results are more accurate.
Furthermore, the terms "first", "second" and "first" 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 one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; either directly or indirectly through intervening media, either internally or in any other relationship. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In the present invention, unless otherwise expressly stated or limited, "above" or "below" a first feature means that the first and second features are in direct contact, or that the first and second features are not in direct contact but are in contact with each other via another feature therebetween. Also, the first feature being "on," "above" and "over" the second feature includes the first feature being directly on and obliquely above the second feature, or merely indicating that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature includes the first feature being directly under and obliquely below the second feature, or simply meaning that the first feature is at a lesser elevation than the second feature.
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 described in this specification can be combined and combined by those skilled in the art.
While the present application has been described in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, apparatus (device), or computer program product. Accordingly, this application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects that may all generally be referred to herein as a "module" or "system. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. A computer program stored/distributed on a suitable medium supplied together with or as part of other hardware, may also take other distributed forms, such as via the Internet or other wired or wireless telecommunication systems.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (devices) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In addition, the embodiment of the invention also provides a display device, and the display device can comprise the display substrate provided by the embodiment. The display device may be: any product or component with a display function, such as an LTPO display device, a Micro LED display device, a liquid crystal panel, electronic paper, an OLED panel, an AMOLED panel, a mobile phone, a tablet computer, a television, a display, a notebook computer, a digital photo frame, and the like.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (10)
1. A deep handwriting authentication method based on a double-tower network is characterized by comprising the following steps:
acquiring a first handwriting image to be authenticated and a second handwriting image of a person to be authenticated;
inputting the first handwriting image into a first input channel of the trained double-tower network model and inputting a second handwriting image into a second input channel of the trained double-tower network model to obtain a recognition result output by the trained double-tower network model; the recognition result comprises: probability of a handwriting belonging to a person to be authenticated;
determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated;
the trained double-tower network model is obtained by training a preset double-tower network model, and is used for carrying out data reconstruction on the first handwriting image and the second handwriting image, extracting first characteristic data from data of the reconstructed first handwriting image, extracting second characteristic data from data of the reconstructed second handwriting image, fusing the first characteristic data and the second characteristic data, and outputting a recognition result based on the fused characteristic data.
2. A method of deep handwriting authentication according to claim 1 and further comprising, before inputting said first handwriting image into a first input channel of a trained two-tower network model and inputting a second handwriting image into a second input channel of said trained two-tower network model:
respectively binarizing the first handwriting image and the second handwriting image;
performing image channel conversion on the binarized first handwriting image and the binarized second handwriting image so as to enable an image channel of the first handwriting image to be matched with an image channel required by the first input channel or an image channel of the second handwriting image to be matched with an image channel required by the second input channel;
wherein the image channel required by the first input channel is the same as the image channel required by the second input channel.
3. The method for deep handwriting authentication according to claim 1, wherein said preset two-tower network model comprises: the system comprises an encoding and decoding network, a backbone network, a feature crossing network and a classification output network, wherein an input channel of the encoding and decoding network comprises: the encoding and decoding network is used for reconstructing data of an input sample image, the backbone network extracts feature data after image information is reconstructed, the feature crossing network is used for performing feature difference calculation on the feature data of the two channel networks of the backbone network, and an identification result is output based on a calculation result of the feature difference;
the characteristic data includes: the first characteristic data and the second characteristic data.
4. A method for deep handwriting authentication according to claim 3 and wherein said encoding and decoding network comprises: a first channel network and a second channel network, the first channel network and the second channel network sharing a weight, the backbone network comprising: the system comprises a third channel network and a fourth channel network, wherein the third channel network and the fourth channel network share weight, the output of the first channel network is connected with the input of the third channel network, the output of the second channel network is connected with the fourth channel network, and the output of the third channel network is connected with the output of the fourth channel network and the input connected with the characteristic cross network.
5. The method for deep handwriting authentication according to claim 4, wherein the step of training the preset two-tower network model comprises:
the method comprises the following steps: constructing a sample set, the sample set comprising: a positive sample pair and a negative sample pair, the positive sample pair and the negative sample pair being a plurality of samples selected from a corresponding data set;
step two: for a sample pair in the sample set, inputting a first sample in the sample pair into the first channel network and inputting a second sample in the sample pair into the second channel network, and obtaining an identification result of the feature crossing network;
the first channel network is configured to transmit a first result of summation of reconstructed data and data of the first sample to the third channel network after data reconstruction is performed on the first sample, the second channel network is configured to transmit a second result of summation of reconstructed data and data of the second sample to the fourth channel network after data reconstruction is performed on the second sample, the third channel network is configured to extract first feature data from the first result and transmit the first feature data to the feature crossing network, the fourth channel network is configured to extract second feature data from the second result and transmit the second feature data to the feature crossing network, the feature crossing network is configured to fuse the first feature data and the second feature data, and an identification result is output based on the fused result;
step three: calculating a contrast loss based on the first result and the second result;
step four: calculating the cross entropy loss of the preset double-tower network model based on the recognition result;
step five: adjusting the weight in the preset double-tower network model based on the contrast loss and the cross entropy loss, and repeating the second step to the fourth step until a training cut-off condition is reached;
step six: and determining the preset double-tower network model reaching the training cut-off condition as a trained double-tower network model.
6. The method for deep handwriting authentication according to claim 5, wherein said step of constructing a sample set comprises:
acquiring a first data set of a plurality of handwriting images of the same person and a second data set of a plurality of handwriting images of different persons;
randomly selecting a preset number of positive sample pairs from the first data set;
randomly selecting a preset number of negative sample pairs from the second data set;
forming a set of samples using the pair of positive samples and the pair of negative samples.
7. A method for deep handwriting authentication according to claim 5 and wherein said step of adjusting weights in said pre-defined two-tower network model based on said contrast loss and said cross-entropy loss comprises:
weighting and summing the contrast loss and the cross entropy loss to obtain a total loss value;
and adjusting the weight in the preset double-tower network model according to the direction of reducing the total loss value.
8. A method for deep handwriting authentication according to claim 7 and wherein said step of adjusting weights in said pre-set two-tower network model in the direction that decreases said total loss value comprises:
and adjusting the weights of the coding and decoding network, the backbone network and the feature crossing network in the preset double-tower network model by using a back propagation algorithm according to the direction of reducing the total loss value.
9. The method for deep handwriting authentication according to claim 1, wherein said determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated comprises:
and when the probability of the handwriting belonging to the person to be authenticated in the recognition result is greater than a preset threshold value, determining that the handwriting in the first handwriting image is the handwriting of the person to be authenticated.
10. A deep handwriting authentication device based on a double-tower network is characterized by comprising:
the system comprises an acquisition module, a verification module and a verification module, wherein the acquisition module is used for acquiring a first handwriting image to be verified and a second handwriting image of a person to be verified;
the recognition module is used for inputting the first handwriting image into a first input channel of the trained double-tower network model and inputting the second handwriting image into a second input channel of the trained double-tower network model to obtain a recognition result output by the trained double-tower network model; the recognition result comprises: probability of a handwriting belonging to a person to be authenticated;
the determining module is used for determining whether the handwriting in the first handwriting image is the handwriting of the person to be authenticated according to the probability of the handwriting belonging to the person to be authenticated;
the trained double-tower network model is obtained by training a preset double-tower network model, and is used for carrying out data reconstruction on the first handwriting image and the second handwriting image, extracting first characteristic data from data of the reconstructed first handwriting image, extracting second characteristic data from data of the reconstructed second handwriting image, fusing the first characteristic data and the second characteristic data, and outputting a recognition result based on the fused characteristic data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010954616.8A CN112215076B (en) | 2020-09-11 | 2020-09-11 | Deep handwriting identification method and device based on double-tower network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010954616.8A CN112215076B (en) | 2020-09-11 | 2020-09-11 | Deep handwriting identification method and device based on double-tower network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112215076A true CN112215076A (en) | 2021-01-12 |
| CN112215076B CN112215076B (en) | 2022-08-12 |
Family
ID=74050406
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010954616.8A Active CN112215076B (en) | 2020-09-11 | 2020-09-11 | Deep handwriting identification method and device based on double-tower network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215076B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610065A (en) * | 2021-10-09 | 2021-11-05 | 北京世纪好未来教育科技有限公司 | Handwriting recognition method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014169837A1 (en) * | 2013-04-18 | 2014-10-23 | 武汉汉德瑞庭科技有限公司 | Method and system for online handwriting authentication on the basis of palm side surface information |
| CN110659599A (en) * | 2019-09-19 | 2020-01-07 | 安徽七天教育科技有限公司 | Scanning test paper-based offline handwriting authentication system and using method thereof |
| CN111046774A (en) * | 2019-12-06 | 2020-04-21 | 国网湖北省电力有限公司电力科学研究院 | Chinese signature handwriting identification method based on convolutional neural network |
| CN111046760A (en) * | 2019-11-29 | 2020-04-21 | 山东浪潮人工智能研究院有限公司 | Handwriting identification method based on domain confrontation network |
-
2020
- 2020-09-11 CN CN202010954616.8A patent/CN112215076B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014169837A1 (en) * | 2013-04-18 | 2014-10-23 | 武汉汉德瑞庭科技有限公司 | Method and system for online handwriting authentication on the basis of palm side surface information |
| CN110659599A (en) * | 2019-09-19 | 2020-01-07 | 安徽七天教育科技有限公司 | Scanning test paper-based offline handwriting authentication system and using method thereof |
| CN111046760A (en) * | 2019-11-29 | 2020-04-21 | 山东浪潮人工智能研究院有限公司 | Handwriting identification method based on domain confrontation network |
| CN111046774A (en) * | 2019-12-06 | 2020-04-21 | 国网湖北省电力有限公司电力科学研究院 | Chinese signature handwriting identification method based on convolutional neural network |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610065A (en) * | 2021-10-09 | 2021-11-05 | 北京世纪好未来教育科技有限公司 | Handwriting recognition method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112215076B (en) | 2022-08-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Do et al. | Forensics face detection from GANs using convolutional neural network | |
| CN107423701B (en) | Face unsupervised feature learning method and device based on generative confrontation network | |
| EP3471023B1 (en) | Offline identity authentication method and apparatus | |
| Lee et al. | Detecting handcrafted facial image manipulations and GAN-generated facial images using Shallow-FakeFaceNet | |
| Ding et al. | Robust face recognition via multimodal deep face representation | |
| US20220092882A1 (en) | Living body detection method based on facial recognition, and electronic device and storage medium | |
| CN112784929B (en) | Small sample image classification method and device based on double-element group expansion | |
| CN113408574B (en) | License plate classification method, license plate classification device and computer readable storage medium | |
| Baek et al. | Generative adversarial ensemble learning for face forensics | |
| Chen et al. | A pornographic images recognition model based on deep one-class classification with visual attention mechanism | |
| Wu et al. | Comparative analysis and application of LBP face image recognition algorithms | |
| CN114677722A (en) | Multi-supervision human face in-vivo detection method integrating multi-scale features | |
| Zeng et al. | Occlusion‐invariant face recognition using simultaneous segmentation | |
| CN114445052A (en) | Intelligent education student attendance big data statistical method and system based on block chain | |
| CN112215076B (en) | Deep handwriting identification method and device based on double-tower network | |
| Wyzykowski et al. | Multiresolution synthetic fingerprint generation | |
| CN113743365A (en) | Method and device for detecting fraudulent behavior in face recognition process | |
| Sharma et al. | A Survey of Detection and Mitigation for Fake Images on Social Media Platforms | |
| Medvedev et al. | Towards facial biometrics for id document validation in mobile devices | |
| Alhumyani et al. | Efficient generation of cancelable face templates based on quantum image Hilbert permutation | |
| CN115620342A (en) | Cross-modal pedestrian re-identification method, system and computer | |
| CN112613341A (en) | Training method and device, fingerprint identification method and device, and electronic device | |
| CN113572913B (en) | Image encryption method, device, medium and electronic equipment | |
| Wright | Lip-based biometric authentication | |
| Lakshminarasimha et al. | Deep Learning Base Face Anti Spoofing-Convolutional Restricted Basis Neural Network Technique |
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 |