CN112434780A - Target object recognition network model, training method thereof and target object recognition method - Google Patents

Target object recognition network model, training method thereof and target object recognition method Download PDF

Info

Publication number
CN112434780A
CN112434780A CN201910788789.4A CN201910788789A CN112434780A CN 112434780 A CN112434780 A CN 112434780A CN 201910788789 A CN201910788789 A CN 201910788789A CN 112434780 A CN112434780 A CN 112434780A
Authority
CN
China
Prior art keywords
network
interference
training
target object
sample
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
Application number
CN201910788789.4A
Other languages
Chinese (zh)
Other versions
CN112434780B (en
Inventor
徐杨柳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Goldway Intelligent Transportation System Co Ltd
Original Assignee
Shanghai Goldway Intelligent Transportation System Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Goldway Intelligent Transportation System Co Ltd filed Critical Shanghai Goldway Intelligent Transportation System Co Ltd
Priority to CN201910788789.4A priority Critical patent/CN112434780B/en
Publication of CN112434780A publication Critical patent/CN112434780A/en
Application granted granted Critical
Publication of CN112434780B publication Critical patent/CN112434780B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a target object recognition network model, a training method thereof and a target object recognition method. The target object identification method comprises the steps of receiving a target object to be identified with interference factors; inputting a target object to be recognized into a trained target object recognition network model; the target object identification network model comprises an interference elimination network and a target object identification network; the interference exclusion network comprises at least one first type interference exclusion network based on a generative countermeasure network and a second type interference exclusion network connected with the first type interference exclusion network; respectively utilizing a first interference elimination network and a second interference elimination network to eliminate interference of different interference factors in a target object to be identified; and identifying the target object to be identified after interference elimination by using a target object identification network. The technical scheme solves the problem that the anti-interference capability is not strong when the target object with interference is identified by the existing target object identification technology.

Description

Target object recognition network model, training method thereof and target object recognition method
[ technical field ] A method for producing a semiconductor device
The invention relates to the technical field of target object recognition based on a neural network, in particular to a target object recognition network model, a training method thereof and a target object recognition method.
[ background of the invention ]
With the rapid development of target object recognition technology, text images can be well recognized with less external interference, such as print scanning, identification photographs, and the like, but errors tend to occur when interference occurs, including rotation, abnormal illumination, blurring, complex colors, and backgrounds. Because the neural network has a certain noise filtering capability, the industry is dedicated to artificially generating more samples to achieve the purpose of further improving the noise filtering capability of the neural network, but the anti-interference capability of the method is also limited.
[ summary of the invention ]
In view of this, embodiments of the present invention provide a target object recognition network model, a training method thereof, and a target object recognition method, so as to solve the problem that the interference resistance is not strong when the existing target object recognition technology recognizes a target object with interference.
In one aspect, an embodiment of the present invention provides a target object identification method, including: receiving a target object to be identified, wherein the target object to be identified has an interference factor; inputting the target object to be recognized into a trained target object recognition network model; wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network; respectively utilizing the first-class interference elimination network and the second-class interference elimination network to eliminate interference of different interference factors in the target object to be identified; and identifying the target object to be identified after interference elimination by using the target object identification network.
Optionally, the interference rejection network is trained by the following method: acquiring training sample pairs, wherein each training sample pair comprises a standard non-interference training sample and an interference training sample; outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network; extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network; comparing the standard feature and the generated feature to obtain a feature error; in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
Optionally, the first type interference rejection network is trained by the following method: outputting a first generated non-interference sample after the interference training sample passes through the first type interference elimination network; extracting a first standard feature from the standard non-interfering training sample and a first generated feature from the first generated non-interfering sample, respectively, using the target object recognition network; comparing the first standard feature and the first generated feature to obtain a first feature error; in the process of training the first-class interference elimination network, the first-class interference elimination network is trained in an auxiliary mode based on the first characteristic error so as to complete training of the first-class interference elimination network.
Optionally, in the process of training the first-class interference rejection network, the training the first-class interference rejection network based on the first characteristic error in an auxiliary manner to complete the training of the first-class interference rejection network includes: alternately training a generating network and a discriminating network of the generating countermeasure network to optimize the generating network and the discriminating network, respectively; assisting in training the generating network based on the first feature error; and taking the trained generated network as the first-class interference elimination network after the training is finished.
Optionally, the second type interference rejection network is trained by the following method: outputting a second generated non-interference sample after the first generated non-interference sample passes through the second type interference elimination network; extracting a second standard feature from the standard non-interfering training sample and a second generated feature from the second generated non-interfering sample, respectively, using the target object recognition network; comparing the second standard feature and the second generated feature to obtain a second feature error; in the process of training the second-class interference elimination network, the second-class interference elimination network is trained in an auxiliary mode based on the second characteristic error so as to complete training of the second-class interference elimination network.
On the other hand, an embodiment of the present invention further provides a target object recognition apparatus, including: the device comprises a to-be-identified object receiving module, a recognition module and a recognition module, wherein the to-be-identified object receiving module is used for receiving a to-be-identified target object, and the to-be-identified target object has interference factors; the object to be recognized processing module is used for inputting the target object to be recognized into a trained target object recognition network model; wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network; the interference elimination module is used for respectively utilizing the first-class interference elimination network and the second-class interference elimination network to eliminate the interference of different interference factors in the target object to be identified; and the target object identification module is used for identifying the target object to be identified after interference elimination by using the target object identification network.
On the other hand, the embodiment of the invention also provides a training method of a target object recognition model, wherein the target object recognition network model comprises an interference elimination network and a target object recognition network;
the training method comprises the following steps: acquiring training sample pairs, wherein each training sample pair comprises a standard non-interference training sample and an interference training sample; outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network; extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network; comparing the standard feature and the generated feature to obtain a feature error; in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
In another aspect, an embodiment of the present invention further provides a target object recognition model, including: an interference rejection network and a target object identification network; wherein the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network; the first type of interference elimination network and the second type of interference elimination network are respectively used for eliminating different interference factors in a target object to be identified; the target object identification network is used for identifying the target object to be identified after interference elimination.
Compared with the prior art, the technical scheme at least has the following beneficial effects:
according to the target object identification method provided by the embodiment of the invention, the utilized target object identification model comprises an interference elimination network and a target object identification network, wherein the interference elimination network comprises at least one first-class interference elimination network based on a generative countermeasure network and a second-class interference elimination network connected with the first-class interference elimination network. The two types of interference elimination networks can eliminate the interference of different interference factors of the target object to be identified layer by layer, so that the target object identification network only needs to identify the target object to be identified with smaller interference factors (namely, smaller noise), and the complexity and the operation amount of the target object identification network are reduced.
Further, when the target object recognition model is trained, in the process of training the interference elimination network, the target object recognition network is utilized to respectively extract standard features from the standard interference-free training sample and extract generated features from the generated interference-free sample, the standard features and the generated features are compared to obtain feature errors, and then the interference elimination network is trained in an auxiliary mode based on the feature errors. When the target object recognition network is trained, the standard non-interference training sample is used for pre-training, and then after the training of the interference elimination network is completed, the pre-trained target object recognition network is trained by the standard non-interference training sample and the interference training sample, so that the training of the target object recognition network is completed. Therefore, in the whole training process of the target object recognition network model, the interference elimination network and the target object recognition network complement each other and are not independently trained, so that the trained target object recognition network model can better recognize the target object.
[ description of the drawings ]
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic flowchart of an embodiment of a target object identification method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a target object recognition model in the target object recognition method of FIG. 1;
FIG. 3 is a schematic flow diagram of a method of training an interference rejection network in the target object recognition model shown in FIG. 2;
FIG. 4A is a schematic flow chart of a method for training a first type of interference rejection network in the target object recognition model shown in FIG. 2;
fig. 4B is a diagram illustrating an embodiment of a training method for the first type of interference rejection network shown in fig. 4A;
FIG. 5A is a flowchart illustrating a method for training a second-class interference rejection network in the target object recognition model shown in FIG. 2;
fig. 5B is a diagram illustrating an embodiment of a training method for the second type of interference rejection network shown in fig. 5A.
FIG. 6 is a flowchart illustrating a method for training a target object recognition network in the target object recognition model shown in FIG. 2 according to an exemplary embodiment;
fig. 7 is a schematic structural diagram of an embodiment of a target object recognition apparatus according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an embodiment of a training apparatus for a target object recognition network model according to an embodiment of the present invention.
[ detailed description ] embodiments
For better understanding of the technical solutions of the present invention, the following detailed descriptions of the embodiments of the present invention are provided with reference to the accompanying drawings.
It should be understood that the described embodiments are only some embodiments of the invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a flowchart illustrating a target object identification method according to an embodiment of the present invention. Referring to fig. 1, the target object identification method includes the steps of:
step 101, receiving a target object to be identified, wherein the target object to be identified has an interference factor.
And 102, inputting the target object to be recognized into a trained target object recognition network model. Wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference rejection network comprises at least one first type of interference rejection network based on a generative countermeasure network and a second type of interference rejection network connected to the first type of interference rejection network.
103, respectively utilizing the first-class interference elimination network and the second-class interference elimination network to eliminate the interference of different interference factors in the target object to be identified.
And 104, identifying the target object to be identified after interference elimination by using the target object identification network.
In this embodiment, a text image (as the target object to be recognized) with an interference factor is recognized, and an application scene of the text on the image is recognized as an example. In this embodiment, the text to be recognized and different interference factors are included in the text picture to be recognized. One type of interference factors includes, but is not limited to, image background interference, illumination interference, blur interference, and color interference, and these interferences may be regarded as noise interference of the text image to be recognized. Another type of interference factor includes, but is not limited to, text skew perspective interference, text warp interference, etc. of the text itself to be recognized.
Fig. 2 is a schematic structural diagram of a target object recognition model in the target object recognition method shown in fig. 1.
Referring to fig. 2, the target object recognition network model 21 includes an interference rejection network 211 and a target object recognition network 212. The interference exclusion network 211 includes at least one first type interference exclusion network 2111 based on a generative countermeasure network and a second type interference exclusion network 2112 connected to the first type interference exclusion network 2111. The first-type interference rejection network 2111 and the second-type interference rejection network 2112 are respectively used for rejecting different interference factors in the target object to be identified. The target object recognition network 212 is configured to perform target object recognition on the target object to be recognized after interference elimination. As shown in fig. 2, the target object to be recognized is a character image containing characters "generator" and background color (black), and after passing through the target object recognition model 21, the characters "generator" on the target object to be recognized can be recognized.
Specifically, the interference elimination network 211 in the target object identification network model 21 provided in this embodiment may eliminate different interference factors on the target object to be identified first, and then the target object identification network 212 in the target object identification network model 21 is used to identify the target object to be identified from which the interference factors are eliminated, so that the target object identification network only needs to identify the target object to be identified with a smaller interference factor, thereby reducing the complexity and the computation workload of the target object identification network.
Further, the interference rejection network 211 includes a first type interference rejection network 2111 and a second type interference rejection network 2112 connected to the first type interference rejection network 2111. The first-type interference rejection network 2111 is an interference rejection network determined based on a generation network G in a Generative Adaptive Networks (GAN). In this embodiment, the first-class interference elimination network 2111 is configured to eliminate noise interference (i.e., any one or more interference factors of image background interference, illumination interference, blur interference, and color interference) in a target object to be identified. The second type of interference rejection network 2112 is an orientation correction network, and for example, a correction network based on a thin plate function model (TPS) or a correction network based on Affine Transformation (Affine Transformation) may be used to perform geometric correction on the target object to be recognized. In this embodiment, the second-class interference elimination network 2112 is configured to eliminate any one or more interference factors of text oblique perspective interference and text bending interference in the target object to be identified.
The following describes in detail the training method of the target object recognition model provided in the present embodiment.
Fig. 3 is a flowchart illustrating a training method of an interference rejection network in the target object recognition model shown in fig. 2. Referring to fig. 3, the training method includes the steps of:
301, acquiring training sample pairs, wherein each training sample pair comprises a standard non-interference training sample and an interference training sample;
step 302, outputting the interference training sample to generate an interference-free sample after the interference training sample passes through the interference elimination network;
step 303, extracting standard features from the standard non-interference training samples and extracting generation features from the generated non-interference samples respectively by using the target object recognition network;
step 304, comparing the standard feature and the generated feature to obtain a feature error;
step 305, in the process of training the interference elimination network, training the interference elimination network based on the characteristic error to complete the training of the interference elimination network.
Different from the prior art, in the embodiment, in the process of training the interference elimination network, the target object recognition network is used for respectively extracting the standard features from the standard interference-free training sample and extracting the generated features from the generated interference-free sample obtained after the interference elimination network passes through, then the standard features and the generated features are compared to obtain the feature errors, and the interference elimination network is trained based on the feature errors in an auxiliary manner, so that the trained interference elimination network has stronger interference elimination capability.
In step 301, the training sample pairs may be generated using existing manual sample generation tools, and each generated training sample pair includes a standard non-interfering training sample and an interfering training sample. The standard non-interference training sample is a target object sample without interference factors, and the interference training sample is a target object sample with different interference factors added on the basis of the standard non-interference training sample.
In step 302, the training samples with interference generated in step 301 are input into the interference elimination network and then output to generate non-interference samples. The interference elimination network is a neural network, and interference elimination can be carried out on input interference training samples through the neural network so as to output interference-free samples.
In the target object identification model provided in this embodiment, two types of interference rejection networks are set for two different types of interference factors, which are respectively a first type of interference rejection network and a second type of interference rejection network. Wherein the first-class interference elimination network is used for eliminating noise interference (for example, any one or more interference factors of background interference, illumination interference, fuzzy interference and color interference) in the interfered training sample. The second type interference elimination network is used for eliminating any one or more interference factors of target object tilting perspective interference and target object bending interference in the interfered training sample. Specific training methods for these two types of interference rejection networks will be described in detail in the examples below.
As described in steps 304 and 305, a target object recognition network is used to extract standard features from the standard non-interfering training samples and generate features from the generated non-interfering training samples, respectively, and the standard features and the generated features are compared to obtain a feature error.
Specifically, the target object recognition network may extract features (such as end points, branch points, concave-convex portions, and the like of characters) from a target object to be recognized, and then perform logical combination judgment according to positions and mutual relations of the extracted features to obtain a recognition result. The target object identification network may adopt an Attention-based identification network or a ctc (connectionist Temporal classification) -based identification network or other target object identification networks derived based on the two frameworks.
In the embodiment, a characteristic extraction part of the target object recognition network is used for extracting standard characteristics from the standard non-interference training sample and extracting generated characteristics from the generated non-interference sample respectively, and then the standard characteristics and the generated characteristics are compared to obtain a characteristic error between the standard characteristics and the generated characteristics. The characteristic error represents the difference between the generated non-interfering sample and a standard non-interfering sample.
In the process of training the interference rejection network, the training of the interference rejection network is assisted based on the characteristic errors, so as to complete the training of the interference rejection network, as in step 306. In other words, in the process of training the interference elimination network, the relevant parameters in the interference elimination network are further optimized and adjusted by using the characteristic error fed back by the target object identification network.
Fig. 4A is a flowchart illustrating a training method of the first-class interference rejection network in the target object recognition model shown in fig. 2. Referring to fig. 4A, the training method includes the steps of:
step 401, outputting a first generated non-interference sample after the interference training sample passes through the first type interference elimination network;
step 402, extracting a first standard characteristic from the standard non-interference training sample and a first generation characteristic from the first generation non-interference sample respectively by using the target object recognition network;
step 403, comparing the first standard feature and the first generated feature to obtain a first feature error;
step 404, in the process of training the first-class interference elimination network, training the first-class interference elimination network based on the first characteristic error in an auxiliary manner so as to complete training of the first-class interference elimination network.
In this embodiment, the first type of interference rejection network is an interference rejection network based on a generative countermeasure network. Those skilled in the art understand that a generative confrontation network (hereinafter referred to as GAN network) is a deep learning model, the GAN network is composed of a generative network G and a discriminating network D, the goal of the generative network G is to generate a picture close to the real as much as possible to deceive the discriminating network D, and the goal of the discriminating network D is to judge whether an input picture is a picture generated by the generative network G or a real picture, that is, the output of the discriminating network represents the probability that the input picture is a real picture, if 1, 100% is a real picture, and if 0, the output represents an impossible picture. Thus, the generation network G and the discrimination network D constitute a dynamic gaming process.
Fig. 4B is a diagram illustrating an embodiment of a training method for the first type of interference rejection network shown in fig. 4A. In fig. 4B, (a) is a schematic diagram of the fixed discrimination network D, training the generation network G; (b) the method is a schematic diagram for fixedly generating parameters of the network G and training the discrimination network D.
As shown in fig. 4B, wherein (a) and (B) show the process of alternately training the generation network and the discriminant network of the generation countermeasure network to optimize the generation network and the discriminant network, respectively.
Specifically, the process (a) of fixing the parameters of the discriminant network D to train the generation network G includes outputting the first generation non-interference sample after the interfering training sample passes through the generation network G, then inputting the first generation non-interference sample and the interfering training sample into the discriminant network D, determining the probability that the first generation non-interference sample output by the discrimination network D is the standard non-interference training sample (i.e., a real non-interference sample), and feeding back the determination result to the generation network G to adjust the parameters of the generation network G.
(b) Shown is a process of fixing the parameters of the generated network G to train the discrimination network D after adjusting the parameters of the generated network G in accordance with the training process shown in (a) above. Randomly selecting one of a first generated non-interfering sample or a standard non-interfering training sample output via the generating network G as an input sample to be input to a discriminating network D, which will compare the interfering training sample with the received input sample in accordance therewith, thereby identifying whether the input sample is the first generated non-interfering sample or the standard non-interfering training sample, to output a probability that the input sample is the standard non-interfering training sample.
The generating network and the discriminating network are alternately trained in the above-described manner to optimize parameters in the generating network and the discriminating network, respectively.
In contrast to the prior art, with continued reference to (a) in fig. 4B, in the process of training the generation network G, the target object recognition network is used to extract a first standard feature from the standard non-interference training sample and a first generation feature from the first generation non-interference sample, and then the first standard feature and the first generation feature are compared to obtain a first feature error, so as to assist in training the generation network based on the first feature error, and the generation network after training is used as the first type of interference elimination network after training is completed.
Further, in the process of training the generation network G, pixel characteristics are extracted from the standard non-interference training sample and the first generation non-interference sample respectively to carry out pixel level comparison on the two samples, so that a pixel level error between the two samples is determined, and the generation network is trained in an auxiliary mode by combining the pixel level error. Therefore, while training the generation network based on the feedback of the discriminant network, the first generation non-interference sample obtained in the training process is compared with the standard non-interference training sample in a pixel level and a feature level, so that the generation network can generate a more detailed first generation non-interference sample.
From the mathematical model point of view, the first type of interference rejection network can be represented by the following formula:
G*＝argminGmaxDLcGAN(G，D)+λ1LL1(G)+λ2LL2(G)
wherein G is a first interference rejection network, LcGAN(G, D) is the generic loss function of the GAN network, LL1(G) Is a loss function of a pixel-level comparison of said first generated non-interfering sample generated with a standard non-interfering training sample, LL2(G) Is a loss function that compares the first generated non-interfering sample with a standard non-interfering training sample at a feature level. During the training process, L is adjusted by adjusting the discrimination network DcGAN(G, D) is as large as possible, and the generating network G is adjusted to LcGAN(G, D) is as small as possible, G forming a countermeasure relationship with D, such that the first generated non-interfering sample is close to a standard non-interfering training sample. L isL1(G) By pixel-level comparison, the first generated non-interfering sample is approximated in detail to a standard non-interfering training sample, LL2(G) And through feature level comparison, the information which has decisive effect on the text recognition result in the first generation non-interference sample is consistent with the standard non-interference training sample.
It should be noted that, in the target object identification model, the number of the first-type interference elimination networks may be determined according to interference factors to be eliminated. For example, a plurality of interference factors (e.g., illumination interference, blur interference, and color interference) may be excluded by one interference exclusion network of the first type, in which case the computation amount of the interference exclusion network of the first type is large and the network structure is complex. For another example, a plurality of the first-type interference elimination networks may be arranged in cascade, and each of the first-type interference elimination networks is used for eliminating one interference factor, in which case, the calculation amount of each of the first-type interference elimination networks is greatly reduced.
Fig. 5A is a flowchart illustrating a method for training a second-class interference rejection network in the target object recognition model shown in fig. 2 according to an embodiment. Referring to fig. 5A, the training method includes the steps of:
step 501, outputting a second generated non-interference sample after the first generated non-interference sample passes through the second type interference elimination network;
step 502, extracting a second standard characteristic from the standard non-interference training sample and extracting a second generation characteristic from the second generation non-interference sample by using the character recognition network;
step 503, comparing the second standard feature and the second generated feature to obtain a second feature error;
step 504, in the process of training the second-class interference elimination network, training the second-class interference elimination network based on the second characteristic error in an auxiliary manner so as to complete the training of the second-class interference elimination network.
In this embodiment, the second type interference elimination network receives the first generated interference-free sample output after passing through the first type interference elimination network. The first generation non-interference sample is a sample subjected to illumination interference, fuzzy interference and color interference elimination, and any one or more of target object oblique perspective interference or target object bending interference of the first generation non-interference sample is further eliminated through the second type interference elimination network.
The second type of interference rejection network is a directional correction network, and for example, a correction network based on a thin plate function model (TPS) or a correction network based on Affine Transformation (Affine Transformation) may be used. The method for training the orientation correction network may be determined according to different correction networks, for example, taking a correction network based on Affine Transformation (Affine Transformation) as an example, a neural network is used to regress four corners of a character, and geometric means such as Affine Transformation are used to correct the character to a horizontal position.
Fig. 5B is a schematic diagram of a specific embodiment of a training method for a second type interference rejection network.
Different from the prior art, in this embodiment, the target object recognition network is used to extract a second standard feature from the standard non-interference training sample and a second generated feature from the second generated non-interference sample, and then the second standard feature and the second generated feature are compared to obtain a second feature error, so as to assist in training the second-class interference rejection network based on the second feature error in the process of training the second-class interference rejection network.
In practical application, for example, the orientation correction network is connected with the target object recognition network, the second generated non-interference sample obtained after the orientation correction network is corrected is directly input into the target object recognition network, and the adjustment required by the current correction is estimated according to the recognition effect of the target object recognition network.
Fig. 6 is a flowchart illustrating a method for training a target object recognition network in the target object recognition model shown in fig. 2 according to an embodiment. Referring to fig. 6, the training method includes the steps of:
601, pre-training the target object recognition network by using the standard non-interference training sample;
step 602, after the training of the interference elimination network is completed, training the pre-trained target object recognition network by using the standard non-interference training sample and the interference training sample to obtain the trained target object recognition network.
In this embodiment, the target object identification network may adopt an identification network based on Attention, an identification network based on ctc (connectionist Temporal classification), or other target object identification networks derived based on these two frames.
The target object recognition network includes a feature extraction portion, an encoding portion (if needed), and a decoding portion.
The training of the target object recognition network is that the trained target object recognition network is expected to substantially conform to the features extracted from the second generated non-interference sample and the standard non-interference training sample, so that the trained target object recognition network can correctly recognize the target object from the target object to be recognized.
To achieve this, the performance of the target object recognition network is not expected to be too strong, and therefore the target object recognition network is first trained (i.e., pre-trained) with standard non-interfering training samples, making it more susceptible to interference. Then, the pre-trained target object recognition network is used for training the interference elimination network (including the first type interference elimination network and the second type interference elimination network) in an auxiliary mode, after the training of the interference elimination network is completed, the pre-trained target object recognition network is trained by using a standard non-interference training sample and an interference training sample, so that the training of the target object recognition network is completed, and the performance of the target object recognition network is further improved.
Therefore, in the whole training process of the target object recognition network model, the interference elimination network and the target object recognition network complement each other and are not independently trained, so that the trained target object recognition network model can better recognize the target object.
Fig. 7 is a schematic structural diagram of an embodiment of a target object recognition apparatus according to an embodiment of the present invention. Referring to fig. 7, the target object recognition apparatus 7 includes: the target object receiving module 71 is configured to receive a target object to be identified, where the target object to be identified has an interference factor. A to-be-recognized object processing module 72, configured to input the to-be-recognized target object into a trained target object recognition network model; wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference rejection network comprises at least one first type of interference rejection network based on a generative countermeasure network and a second type of interference rejection network connected to the first type of interference rejection network. And the interference elimination module 73 is configured to perform interference elimination on different interference factors in the target object to be identified by using the first-class interference elimination network and the second-class interference elimination network, respectively. And the target object identification module 74 is configured to identify the target object to be identified after interference elimination by using the target object identification network.
The target object recognition means 7 further comprise an interference rejection network training module 75. The interference rejection network training module 75 is configured to obtain training sample pairs, where each training sample pair includes a standard non-interference training sample and an interference training sample; outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network; extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network; comparing the standard feature and the generated feature to obtain a feature error; in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
The interference rejection network training module 75 includes a first type interference rejection network training unit 751, configured to output a first generated interference-free sample after the interference-free training sample passes through the first type interference rejection network; extracting a first standard feature from the standard non-interfering training sample and a first generated feature from the first generated non-interfering sample, respectively, using the target object recognition network; comparing the first standard feature and the first generated feature to obtain a first feature error; in the process of training the first-class interference elimination network, the first-class interference elimination network is trained in an auxiliary mode based on the first characteristic error so as to complete training of the first-class interference elimination network.
The first type interference rejection network training unit 751 is further configured to alternately train a generation network and a discrimination network of the generation countermeasure network to optimize the generation network and the discrimination network, respectively; assisting in training the generating network based on the first feature error; and taking the trained generated network as the first-class interference elimination network after the training is finished.
The interference elimination network training module 75 includes a second type interference elimination network training unit 752, configured to output a second generated non-interference sample after the first generated non-interference sample passes through the second type interference elimination network; extracting a second standard feature from the standard non-interfering training sample and a second generated feature from the second generated non-interfering sample, respectively, using the target object recognition network; comparing the second standard feature and the second generated feature to obtain a second feature error; in the process of training the second-class interference elimination network, the second-class interference elimination network is trained in an auxiliary mode based on the second characteristic error so as to complete training of the second-class interference elimination network.
Fig. 8 is a schematic structural diagram of an embodiment of a training apparatus for a target object recognition network model according to an embodiment of the present invention. Referring to fig. 8, the training device 8 includes: a training sample obtaining module 81, configured to obtain training sample pairs, where each training sample pair includes a standard non-interference training sample and an interference training sample. And an interference-free sample generation module 82, configured to output the interference-free training sample after passing through the interference elimination network. And a generating characteristic extracting module 83, configured to extract a standard characteristic from the standard non-interference training sample and a generating characteristic from the generating non-interference training sample, respectively, by using the target object recognition network. A feature error determination module 84 for comparing the standard feature and the generated feature to obtain a feature error. And an auxiliary training module 85, configured to, in the process of training the interference rejection network, assist in training the interference rejection network based on the characteristic error, so as to complete training of the interference rejection network.
Wherein the interference rejection network comprises at least one first type of interference rejection network based on a generative countermeasure network. The generate no-interference sample module 82 is configured to output a first generate no-interference sample after the interfering training sample passes through the first type interference elimination network. The generation characteristic extraction module 83 is configured to extract a first standard characteristic from the standard non-interfering training sample and a first generation characteristic from the first generation non-interfering training sample, respectively, by using the target object recognition network. The feature error determination module 84 is configured to compare the first standard feature and the first generated feature to obtain a first feature error. The auxiliary training module 85 is configured to, in the process of training the first class interference elimination network, assist in training the first class interference elimination network based on the first characteristic error, so as to complete training of the first class interference elimination network.
The interference rejection network further comprises a second type of interference rejection network connected to the first type of interference rejection network. The generate non-interference sample module 82 is further configured to output a second generate non-interference sample after the first generate non-interference sample passes through the second type interference rejection network. The generated feature extraction module 83 is further configured to extract a second standard feature from the standard non-interfering training sample and a second generated feature from the second generated non-interfering training sample, respectively, using the target object recognition network. The feature error determination module 84 is further configured to compare the second standard feature and the second generated feature to obtain a second feature error. The auxiliary training module 85 is further configured to assist in training the second-class interference elimination network based on the second characteristic error in the process of training the second-class interference elimination network, so as to complete training of the second-class interference elimination network.
An embodiment of the present invention further provides a computer-readable storage medium, where the storage medium stores a computer program, and the computer program is configured to execute each step in the above-described embodiment of the target object identification method.
An embodiment of the present invention further provides an electronic device, where the electronic device includes: a processor; a memory for storing the processor-executable instructions; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the steps in the embodiment of the target object identification method.
An embodiment of the present invention further provides a computer-readable storage medium, where the storage medium stores a computer program, and the computer program is configured to execute each step in the embodiment of the method for training a target object recognition network model.
An embodiment of the present invention further provides an electronic device, where the electronic device includes: a processor; a memory for storing the processor-executable instructions; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the steps in the embodiment of the training method for the target object recognition network model.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (16)

1. A target object recognition method, comprising:
receiving a target object to be identified, wherein the target object to be identified has an interference factor;
inputting the target object to be recognized into a trained target object recognition network model; wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network;
respectively utilizing the first-class interference elimination network and the second-class interference elimination network to eliminate interference of different interference factors in the target object to be identified;
and identifying the target object to be identified after interference elimination by using the target object identification network.
2. The method of claim 1, wherein the interference rejection network is trained by:
acquiring training sample pairs, wherein each training sample pair comprises a standard non-interference training sample and an interference training sample;
outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network;
extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network;
comparing the standard feature and the generated feature to obtain a feature error;
in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
3. The method of claim 2, wherein the first type of interference rejection network is trained by:
outputting a first generated non-interference sample after the interference training sample passes through the first type interference elimination network;
extracting a first standard feature from the standard non-interfering training sample and a first generated feature from the first generated non-interfering sample, respectively, using the target object recognition network;
comparing the first standard feature and the first generated feature to obtain a first feature error;
in the process of training the first-class interference elimination network, the first-class interference elimination network is trained in an auxiliary mode based on the first characteristic error so as to complete training of the first-class interference elimination network.
4. The method of claim 3, wherein the assisting in training the first class of interference rejection network based on the first characteristic error in training the first class of interference rejection network to complete the training of the first class of interference rejection network comprises:
alternately training a generating network and a discriminating network of the generating countermeasure network to optimize the generating network and the discriminating network, respectively;
assisting in training the generating network based on the first feature error;
and taking the trained generated network as the first-class interference elimination network after the training is finished.
5. The method of claim 3, wherein the second type of interference rejection network is trained by:
outputting a second generated non-interference sample after the first generated non-interference sample passes through the second type interference elimination network;
extracting a second standard feature from the standard non-interfering training sample and a second generated feature from the second generated non-interfering sample, respectively, using the target object recognition network;
comparing the second standard feature and the second generated feature to obtain a second feature error;
in the process of training the second-class interference elimination network, the second-class interference elimination network is trained in an auxiliary mode based on the second characteristic error so as to complete training of the second-class interference elimination network.
6. A target object recognition apparatus, comprising:
the device comprises a to-be-identified object receiving module, a recognition module and a recognition module, wherein the to-be-identified object receiving module is used for receiving a to-be-identified target object, and the to-be-identified target object has interference factors;
the object to be recognized processing module is used for inputting the target object to be recognized into a trained target object recognition network model; wherein the target object identification network model comprises an interference rejection network and a target object identification network; the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network;
the interference elimination module is used for respectively utilizing the first-class interference elimination network and the second-class interference elimination network to eliminate the interference of different interference factors in the target object to be identified;
and the target object identification module is used for identifying the target object to be identified after interference elimination by using the target object identification network.
7. The apparatus of claim 6, further comprising an interference rejection network training module: the interference elimination network training module is used for acquiring training sample pairs, and each training sample pair comprises a standard non-interference training sample and an interference training sample; outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network; extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network; comparing the standard feature and the generated feature to obtain a feature error; in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
8. A training method of a target object recognition network model is characterized in that the target object recognition network model comprises an interference elimination network and a target object recognition network;
the training method comprises the following steps:
acquiring training sample pairs, wherein each training sample pair comprises a standard non-interference training sample and an interference training sample;
outputting the interference-free training sample to generate an interference-free sample after the interference-free training sample passes through the interference elimination network;
extracting standard features from the standard non-interfering training samples and generating features from the generating non-interfering samples respectively by using the target object recognition network;
comparing the standard feature and the generated feature to obtain a feature error;
in the process of training the interference elimination network, the interference elimination network is trained in an auxiliary mode based on the characteristic errors so as to complete the training of the interference elimination network.
9. The method of claim 8, wherein the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network;
the outputting the interference-free training sample after passing through the interference elimination network to generate an interference-free sample comprises: outputting a first generated non-interference sample after the interference training sample passes through the first type interference elimination network;
the extracting, with the target object recognition network, the standard features from the standard non-interfering training samples and the generating features from the generating non-interfering samples, respectively, comprises: extracting a first standard feature from the standard non-interfering training sample and a first generated feature from the first generated non-interfering sample, respectively, using the target object recognition network;
the comparing the standard features and the generated features to obtain feature errors comprises: comparing the first standard feature and the first generated feature to obtain a first feature error;
in the process of training the interference elimination network, the training the interference elimination network based on the characteristic error to complete the training of the interference elimination network includes: in the process of training the first-class interference elimination network, the first-class interference elimination network is trained in an auxiliary mode based on the first characteristic error so as to complete training of the first-class interference elimination network.
10. The method of claim 9, wherein the interference rejection network further comprises a second type of interference rejection network connected to the first type of interference rejection network;
the outputting the interference-free training sample after passing through the interference elimination network to generate an interference-free sample comprises: outputting a second generated non-interference sample after the first generated non-interference sample passes through the second type interference elimination network;
the extracting, with the target object recognition network, the standard features from the standard non-interfering training samples and the generating features from the generating non-interfering samples, respectively, comprises: extracting a second standard feature from the standard non-interfering training sample and a second generated feature from the second generated non-interfering sample, respectively, using the target object recognition network;
the comparing the standard features and the generated features to obtain feature errors comprises: comparing the second standard feature and the second generated feature to obtain a second feature error;
in the process of training the interference elimination network, the training the interference elimination network based on the characteristic error to complete the training of the interference elimination network includes: in the process of training the second-class interference elimination network, the second-class interference elimination network is trained in an auxiliary mode based on the second characteristic error so as to complete training of the second-class interference elimination network.
11. A training side device of a target object recognition network model is characterized in that the target object recognition network model comprises an interference elimination network and a target object recognition network;
the training apparatus includes:
the training sample acquisition module is used for acquiring training sample pairs, and each training sample pair comprises a standard non-interference training sample and an interference training sample;
the interference-free sample generation module is used for outputting the interference-free training sample after the interference elimination network;
the generated feature extraction module is used for extracting standard features from the standard non-interference training samples and extracting generated features from the generated non-interference samples by using the target object recognition network;
a feature error determination module for comparing the standard feature and the generated feature to obtain a feature error;
and the auxiliary training module is used for training the interference elimination network in an auxiliary mode based on the characteristic errors in the process of training the interference elimination network so as to finish the training of the interference elimination network.
12. A target object recognition network model, comprising: an interference rejection network and a target object identification network; wherein the interference exclusion network comprises at least one first type of interference exclusion network based on a generative countermeasure network and a second type of interference exclusion network connected with the first type of interference exclusion network;
the first type of interference elimination network and the second type of interference elimination network are respectively used for eliminating different interference factors in a target object to be identified; the character recognition network is used for recognizing the target object to be recognized after interference elimination.
13. A computer-readable storage medium storing a computer program for executing the target object identifying method according to any one of claims 1 to 5.
14. An electronic device, the electronic device comprising:
a processor;
a memory for storing the processor-executable instructions;
the processor is used for reading the executable instructions from the memory and executing the instructions to realize the target object identification method of any one of the claims 1 to 5.
15. A computer-readable storage medium, storing a computer program for executing the method for training the target object recognition network model according to any one of claims 8 to 10.
16. An electronic device, the electronic device comprising:
a processor;
a memory for storing the processor-executable instructions;
the processor is configured to read the executable instructions from the memory and execute the instructions to implement the training method of the target object recognition network model according to any one of claims 8 to 10.
CN201910788789.4A 2019-08-26 2019-08-26 Target object recognition network model, training method thereof and target object recognition method Active CN112434780B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910788789.4A CN112434780B (en) 2019-08-26 2019-08-26 Target object recognition network model, training method thereof and target object recognition method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910788789.4A CN112434780B (en) 2019-08-26 2019-08-26 Target object recognition network model, training method thereof and target object recognition method

Publications (2)

Publication Number Publication Date
CN112434780A true CN112434780A (en) 2021-03-02
CN112434780B CN112434780B (en) 2023-05-30

Family

ID=74690475

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910788789.4A Active CN112434780B (en) 2019-08-26 2019-08-26 Target object recognition network model, training method thereof and target object recognition method

Country Status (1)

Country Link
CN (1) CN112434780B (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106548159A (en) * 2016-11-08 2017-03-29 中国科学院自动化研究所 Reticulate pattern facial image recognition method and device based on full convolutional neural networks
CN108280811A (en) * 2018-01-23 2018-07-13 哈尔滨工业大学深圳研究生院 A kind of image de-noising method and system based on neural network
CN108492258A (en) * 2018-01-17 2018-09-04 天津大学 A kind of radar image denoising method based on generation confrontation network
CN108765319A (en) * 2018-05-09 2018-11-06 大连理工大学 A kind of image de-noising method based on generation confrontation network
CN108875486A (en) * 2017-09-28 2018-11-23 北京旷视科技有限公司 Recongnition of objects method, apparatus, system and computer-readable medium
CN109087269A (en) * 2018-08-21 2018-12-25 厦门美图之家科技有限公司 Low light image Enhancement Method and device
CN109360156A (en) * 2018-08-17 2019-02-19 上海交通大学 Single image rain removing method based on the image block for generating confrontation network
US20190220977A1 (en) * 2018-01-16 2019-07-18 Siemens Healthcare Gmbh Cross-Domain Image Analysis and Cross-Domain Image Synthesis Using Deep Image-to-Image Networks and Adversarial Networks
US20190251401A1 (en) * 2018-02-15 2019-08-15 Adobe Inc. Image composites using a generative adversarial neural network

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106548159A (en) * 2016-11-08 2017-03-29 中国科学院自动化研究所 Reticulate pattern facial image recognition method and device based on full convolutional neural networks
CN108875486A (en) * 2017-09-28 2018-11-23 北京旷视科技有限公司 Recongnition of objects method, apparatus, system and computer-readable medium
US20190220977A1 (en) * 2018-01-16 2019-07-18 Siemens Healthcare Gmbh Cross-Domain Image Analysis and Cross-Domain Image Synthesis Using Deep Image-to-Image Networks and Adversarial Networks
CN108492258A (en) * 2018-01-17 2018-09-04 天津大学 A kind of radar image denoising method based on generation confrontation network
CN108280811A (en) * 2018-01-23 2018-07-13 哈尔滨工业大学深圳研究生院 A kind of image de-noising method and system based on neural network
US20190251401A1 (en) * 2018-02-15 2019-08-15 Adobe Inc. Image composites using a generative adversarial neural network
CN108765319A (en) * 2018-05-09 2018-11-06 大连理工大学 A kind of image de-noising method based on generation confrontation network
CN109360156A (en) * 2018-08-17 2019-02-19 上海交通大学 Single image rain removing method based on the image block for generating confrontation network
CN109087269A (en) * 2018-08-21 2018-12-25 厦门美图之家科技有限公司 Low light image Enhancement Method and device

Also Published As

Publication number Publication date
CN112434780B (en) 2023-05-30

Similar Documents

Publication Publication Date Title
CN108229526B (en) Network training method, network training device, image processing method, image processing device, storage medium and electronic equipment
JP7490141B2 (en) IMAGE DETECTION METHOD, MODEL TRAINING METHOD, IMAGE DETECTION APPARATUS, TRAINING APPARATUS, DEVICE, AND PROGRAM
CN111401372B (en) Method for extracting and identifying image-text information of scanned document
Chierchia et al. Guided filtering for PRNU-based localization of small-size image forgeries
US10558841B2 (en) Method and apparatus for recognizing fingerprint ridge point
CN109086714A (en) Table recognition method, identifying system and computer installation
CN111368666B (en) Living body detection method based on novel pooling and attention mechanism double-flow network
CN110543846A (en) Multi-pose face image obverse method based on generation countermeasure network
CN111753595A (en) Living body detection method and apparatus, device, and storage medium
CN111340716B (en) Image deblurring method for improving double-discrimination countermeasure network model
CN109284683A (en) Feature extraction and matching and template renewal for biological identification
CN113191969A (en) Unsupervised image rain removing method based on attention confrontation generation network
Lu et al. Robust blur kernel estimation for license plate images from fast moving vehicles
CN110674759A (en) Monocular face in-vivo detection method, device and equipment based on depth map
CN113592776A (en) Image processing method and device, electronic device and storage medium
CN111783761A (en) Certificate text detection method and device and electronic equipment
CN111523497A (en) Face correction method and device and electronic equipment
CN113807237B (en) Training of in vivo detection model, in vivo detection method, computer device, and medium
CN117496019B (en) Image animation processing method and system for driving static image
CN116596891B (en) Wood floor color classification and defect detection method based on semi-supervised multitasking detection
CN113379611B (en) Image processing model generation method, processing method, storage medium and terminal
CN104966271B (en) Image de-noising method based on biological vision receptive field mechanism
CN112434780B (en) Target object recognition network model, training method thereof and target object recognition method
CN113014914B (en) Neural network-based single face-changing short video identification method and system
CN110134924A (en) Overlay text component extracting method and device, text recognition system and storage medium

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