CN115482595B - Specific character visual sense counterfeiting detection and identification method based on semantic segmentation - Google Patents
Specific character visual sense counterfeiting detection and identification method based on semantic segmentation Download PDFInfo
- Publication number
- CN115482595B CN115482595B CN202211188905.7A CN202211188905A CN115482595B CN 115482595 B CN115482595 B CN 115482595B CN 202211188905 A CN202211188905 A CN 202211188905A CN 115482595 B CN115482595 B CN 115482595B
- Authority
- CN
- China
- Prior art keywords
- picture
- semantic segmentation
- face
- data set
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
- G06V40/40—Spoof detection, e.g. liveness detection
-
- 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
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Software Systems (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a specific character visual sense counterfeiting detection and identification method based on semantic segmentation, belongs to the technical field of depth counterfeiting and detection, and provides a counterfeiting video detection mode which takes depth counterfeiting video detection of specific characters as a research target, constructs a personal characteristic model of a target task based on a basic method of semi-supervision and semantic segmentation, selects and classifies attribute masks of constructed human face regions, and outputs results by integrating classification weights of various attributes. Firstly, constructing a mask data set of a target character region based on semantic segmentation; and secondly, establishing a personal semantic model for visual forgery detection and identification to detect a deep forged video. In the process of manufacturing the data set, the data set is amplified by using a semi-supervised machine learning algorithm, so that the problem of insufficient data set of the specific person is solved, and the manual labeling cost is reduced.
Description
Technical Field
The invention belongs to the technical field of deep forgery and detection, and relates to a method for detecting a forged video, in particular to a method for detecting and identifying visual forgery of a specific character based on semantic segmentation.
Background
Deep forgery is translated by the word "Deep Fake", which is a combination of "Deep learning" and "Fake", i.e., a combination of Deep learning and forgery. Depth forgery is a technology based on deep learning, and refers to making fake videos and images by exchanging faces of people. The term deepFake originates from a machine learning algorithm published by Reddit users deepfakes at 2017 and states that the algorithm can help him to convert a celebrity face into pornographic video. The algorithm is popular with people and media once released, and a hot tide of the research of the visual depth counterfeiting algorithm follows. In 2018, buzzFeed released a section of deep-forged video about a speech released by barake obama, which was made using FakeApp software manufactured by redpit users. From 2017 to 2020, the depth counterfeiting related papers are increased from 3 original papers to more than 250, meanwhile, fakeApp, faceawap, zao, faceApp and other popular fast depth counterfeiting software which can realize no technical cost are also developed in sequence, and various types of counterfeiting videos made by the visual depth counterfeiting technology also cause the worry of people about identity theft, counterfeiting and the propagation of false information on social media.
Currently, the existing visual depth counterfeiting methods can be roughly divided into three types: synthesizing new face, face decoration and face interchange. The new face synthesis means that a non-existent face image is created by using GAN; facial modification refers to modification of certain parts of the original human face; face interchange refers to the exchange of two faces, either partially or wholly.
The method for synthesizing new face completely creates the whole non-existing face image by using the powerful generation countermeasure network GAN, the database of the existing technology for synthesizing new face is created based on the architecture of ProGAN and StyleGAN, and each created fake image carries the specific GAN fingerprint. The face modification method mainly adds some face modifications to the target human face, such as changing hair color or skin color, modifying the gender of the target person, adding glasses to the target person, and the like, and the method also needs to be based on generating the confrontation network GAN, and the current latest StarGAN technology can simultaneously divide the face into a plurality of fields and perform modification operation on the face. The face interchange method is composed of two parts, the first method is to use the face of another person to replace the face of a target person in a video, which is the most popular method in the current visual depth counterfeiting direction, such as methods that deep fakes and FaceSwap are both utilized, and the method can be used for the synthesis of depth counterfeiting videos, unlike the former two methods, which place face synthesis operation on images; the second way is facial expression exchange, also known as facial reconstruction, i.e. the substitution of a facial expression on another face to the facial expression of the target person, such as by changing the expression and actions of obama to complete a fake "speech".
The visual depth counterfeiting detection technology is mainly carried out by the steps of feature extraction, model establishment, detection classification and the like. Firstly, a researcher preprocesses image or video data to be detected, and determines characteristics to be detected according to priori knowledge or image processing means. And then, designing a corresponding algorithm to extract the determined characteristics, and establishing a network model matched with the detection task. And finally, testing the performance of the detection algorithm by using the data to be detected, thereby verifying the scientificity of the selected features and the effectiveness of the classification model. The key for determining the detection performance lies in how to select the relevant features that can effectively distinguish the true and false faces, and how to establish a model with good classification effect.
Different depth forgery detection methods are embodied in different emphasis points in the detection algorithm flow, so that the detection methods can be classified:
the visual depth forgery detection technology based on specific artifacts focuses on detecting a feature determination part in a flowchart, and from the viewpoint of image processing, abnormal phenomena such as blurring, shaking, and ghost existing in a generated image or video are captured at pixel level granularity. The degree of discrimination of artifact characteristics directly influences the performance of the detection algorithm.
The visual depth counterfeiting detection technology based on data driving focuses on a model building part in a detection flow chart, and a well-designed neural network is used for training and classifying time domain and frequency domain information in an extracted counterfeit. The excellent network design can more effectively extract potential subtle features.
The visual depth forgery detection technology based on information inconsistency focuses on capturing inconsistent parts between a forged product and an objective rule from high-level semantics such as biological inherent characteristics, time continuity, motion vectors and the like. Because the extraction process of the high-level semantic features is complex, the technology focuses on two parts of feature determination and feature extraction in the detection flow chart.
Because a specific character has a large amount of available real face data, a large amount of training is carried out by utilizing the generated countermeasure network GAN according to the real face of the specific character, a very vivid deep forged face can be manufactured, meanwhile, the counterfeiting technologies such as Wav2Lip and the like are assisted, so that the deep forged products of the specific character are easily influenced, and the conventional wide-field counterfeiting detection method is not enough to identify the forged products of the specific character with good performance, so that the research of deep counterfeiting detection aiming at the specific character is needed.
Disclosure of Invention
Aiming at the problems, the invention provides a specific character visual counterfeiting detection and identification method based on semantic segmentation, which effectively improves the counterfeiting detection and identification capability.
The invention discloses a specific character visual counterfeiting detection and identification method based on semantic segmentation, which is divided into a semantic segmentation part and a counterfeiting detection and identification part.
The semantic segmentation part performs semantic segmentation on the deeply forged face: labeling the target character image according to eleven characteristics of the human face to form an initial training set; and generating a mask data set of the target person by using the initial training set and adopting a semi-supervised semantic segmentation model.
The forgery detection and identification part performs dot multiplication on the mask data of the target person segmented according to the semantics and the face picture corresponding to the mask data to acquire a specified picture attribute region, and further performs model construction on the acquired region attribute, specifically:
and for the original face picture z of each target person, combining a picture mask a which is respectively divided by five sense organs in a mask data set with a manually selected interested five sense organ region vector V to obtain a facial five sense organ interested region, and then performing point multiplication on the facial five sense organ interested region and the corresponding original face picture to generate a condition tensor T of the required facial interested region.
And performing point multiplication on the input picture z and the tensor T corresponding to the picture z, and inputting the point multiplication processing result p (z) into the generated countermeasure network to perform posture-independent identification processing. The formula of dot product processing of the tensor T of the selected region of interest and the original image z is as follows:
p(z)＝z·T＝z·a·V
inputting p (z) and a given gesture into a generator G; and the generator G generates a corresponding false picture by using a given posture, the discriminator D judges the posture and the identity of the generated false picture, and the confrontation training is continuously carried out until a critical state that the discriminator D considers that the identity of the false picture generated by the generator G is the same as that of the original input picture is reached, so that a face picture irrelevant to the posture is obtained.
After the pose-independent recognition processing, inputting each face segmentation attribute region of the face picture x subjected to pose change into a single convolutional neural network for classification processing, and constructing a new CNN binary classification classifier. The image features are learned through a convolutional network, the output dimensionality is reduced through a pooling layer, the depth features are fused through a full-link layer, a classification result is finally formed and output, and the purpose of identifying whether the input image is a positive sample or a negative sample is achieved.
The invention has the advantages that:
1. the specific character vision counterfeiting detection and identification method based on semantic segmentation constructs a facial mask data set based on specific characters, and can expand the data of face counterfeiting under the condition of not increasing the manual labeling cost.
2. The specific character visual sense counterfeiting detection and identification method based on semantic segmentation can effectively utilize an attention mechanism, so that the detection accuracy of a classifier on a counterfeit sample is greatly improved.
3. The method for detecting and identifying the visual counterfeiting of the specific character based on the semantic segmentation constructs a posture-independent module, realizes the detection of the input pictures in various postures and increases the robustness of the counterfeiting detection.
4. The specific character visual counterfeiting detection and identification method based on semantic segmentation can detect pictures and videos generated by various fake face counterfeiting technologies, and increases the generalization of counterfeiting detection.
Drawings
FIG. 1 is a flow chart of the method for detecting and identifying visual falsification of specific characters based on semantic segmentation according to the present invention;
FIG. 2 is a semantic segmentation network structure.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
The invention discloses a specific character visual sense counterfeiting detection and identification method based on semantic segmentation, which comprises the following specific steps as shown in figure 1:
step 1: and generating a mask data set of the facial feature segmentation of the specific person based on a semantic segmentation method.
101. And collecting a large amount of data of the target person to form an initial training set.
The data set needs to be collected including the following four parts: real human face of the target person, real human face of other persons than the target person, deep fake human face of the target person, and imitators and performers of the target person. Thereby building the basic resources for model training.
In consideration of subsequent processing of video data, the real face of the target person can select a video with the front face of the target person facing the shot as much as possible and with higher definition from the public video website, and the video is downloaded. The real face of the target person is selected as a target collection object, and about 60 hours of video is collected to ensure that the model can effectively learn the real face characteristics of the target person. Collecting the imitators and players of the target person and the real faces of the non-target persons by referring to the real face collecting mode of the target person; the deep forged face of the target person is manufactured by three forging methods of FaceSwap, wav2lip and First Order Motion for the real face of the target person. The real face of the non-target person can be selected with facial features similar to those of the target person as much as possible, so that the discrimination capability of the model on the similar faces in the real scene is enhanced.
And selecting the target person as a protected target object, wherein the video data set of the real human face of the target person is used as a positive sample, and the imitators and performers of the target person, the deep forged human face of the target person and the forged product data set of the non-target person are used as negative samples.
102. And performing semantic segmentation on the face of the collected real face data set and counterfeit product data set of the target person.
For the faces of all positive samples and negative samples, selecting N pictures which are randomly framed in the video, and manually labeling the pictures by using a LabelMe labeling tool, as shown in FIG. 1, wherein in the manual labeling process, the region where 11 parts are located needs to be labeled, and the labeling is as follows: left eye, right eye, left eyebrow, right eyebrow, nose, upper lip, lower lip, hair, left ear, right ear, and neck. And then processing the json format file generated by labeling and the labeled picture original image together to obtain mask data sets with different facial type labels.
103. And training a machine labeling method by constructing a semantic segmentation network Deeplabv3+ of semi-supervised machine learning by adopting the picture in the mask data set manually labeled in the previous step.
Selecting a semantic segmentation network Deeplabv3+ for training, inputting N mask data set pictures generated by manual labeling in the step 102, and performing automatic machine labeling on M unmarked pictures in the positive and negative samples collected in the step 1 through a deep learning model to realize labeling of semi-supervised machine learning, wherein the method comprises the following steps:
A. as shown in FIG. 2, two semantic segmentation networks P with the same structure and different initial weight values are constructed 1 And P 2 ：
P 1 ＝f(X；θ 1 )
P 2 ＝f(X；θ 2 )
P 1 And P 2 For the two semantic segmentation networks Deeplabv3+, the two are different only in initial value of the weight parameter.
Wherein, X represents the input picture after data enhancement is carried out on N marked pictures; theta 1 And theta 2 Respectively represent P 1 And P 2 Weights of the two networks; y represents a pseudo label obtained by two semantic segmentation networks, namely a primary segmentation result. Wherein two semantically split networks P 1 And P 2 Using deplabv 3+, the classifiers Y1 and Y2 use a ResNet101 network. The Deeplabv3+ enables the former layer in the network to encode the context information with different scales by convolution or pooling of input features relative to other semantic segmentation network structures, and enables the latter layer in the network to capture clear object boundaries by gradually replying spatial information, thereby being applicable to the semantic segmentation of the face. For two divided networks, obtaining corresponding one-hot label Y through argmax operation 1 And Y 2 . Then using the two pseudo labels as supervision signals and using Y 2 As P 1 Supervision of (A), Y 1 As P 2 And (4) monitoring, and using cross entropy loss function constraint to improve the performance of the semantic segmentation network.
And finally combining the M marked pictures generated by the semantic segmentation network machine with the N originally manually marked pictures to form a Mask data set.
Step 2: establishing personal semantic model for visual counterfeit detection and identification
The invention constructs a method for detecting and identifying visual sense forgery of a specific character, which is divided into data preprocessing, gesture-independent identification and model construction according to the flow; the method comprises the following specific steps:
201: raw target face data preprocessing
Marking the original face picture of the target person collected in the step 101 as Z ∈ Z, wherein Z is all face data collected in the step 101 and is a face set including a positive sample and a negative sample thereof, for each face picture, combining the picture Mask a obtained by dividing each of five sense organs in the Mask data set obtained in the step 1 with a manually selected five sense organ region vector V of interest (for example, a nose, an eye and an ear respectively represent a vector, and the selected nose, namely the vector V is [1,0,0 ]) to obtain a facial five sense organ region of interest, and then performing point multiplication on the facial five sense organ region of interest and the corresponding original face picture to generate a condition tensor T of the required facial region of interest.
And performing point multiplication on the input picture z and the tensor T corresponding to the picture z, and inputting the point multiplication processing result p (z) into the generated countermeasure network to perform posture-independent identification processing. The formula of dot product processing of the tensor T of the selected region of interest and the original image z is as follows:
p(z)＝z·T＝z·a·V
202: gesture-independent recognition
Inputting a tensor T point multiplication result p (z) corresponding to each picture z and the picture z and a given posture into a generator G; the pose is set to be the direction of the front face of the human face facing the camera, and the angle between the human face and the right front is 0 degrees. The generator G generates a corresponding false picture by using a given posture, judges the posture and the identity of the generated false picture by using the discriminator D, and continuously performs countermeasure training until a critical state that the discriminator D considers that the false picture generated by the generator G is the same as the identity of the original input picture is reached, so that a face picture irrelevant to the posture is obtained.
203: deep forgery detection model construction
After the gesture-independent recognition processing, each face segmentation attribute region of the face picture x subjected to the gesture change is input into a single convolutional neural network for classification processing, and a new CNN two-classification classifier is constructed. Wherein, the picture characteristics are learned through a convolutional network, and the output dimensionality is reduced through a pooling layer. And the depth features are fused through the full connection layer, and finally a classification result is formed and output, so that the purpose of identifying whether the input picture is a positive sample or a negative sample is achieved.
Wherein, in the design of the loss function, the classification loss is measured by a Binary Cross Entropy (BCE) loss function, L Y Is defined formally as the following equation:
L Y (x，y)＝BCE(p，y)＝-(y*log(p))+(1-y)*log(1-p)
and p is the predicted classification output of the classifier, y is epsilon {0,1} is a true label and a sigmoid activation function is adopted to process the output in the two classification tasks.
In summary, the method for detecting and identifying the visual sense forgery of the specific character based on the semantic segmentation establishes a mask training data set of the facial image segmentation of the specific character, utilizes a semi-supervised machine learning algorithm to amplify the data set, solves the problem of insufficient data set of the specific character and reduces the manual labeling cost; a semantic segmentation module and a posture change module are added before a two-classification detection model, different feature modeling and fitting modes are adopted, the counterfeit detection and identification capabilities are improved, and the accuracy of the two-classification detection model can be improved by 5-10 percentage points.
Claims (3)
1. A specific character visual sense forgery detection and identification method based on semantic segmentation is characterized in that: the method comprises the following steps of dividing into a semantic division part and a forgery detection and identification part;
the semantic segmentation part performs semantic segmentation on the deeply forged face: labeling the target figure image according to eleven features of the human face to form an initial training set; generating a mask data set of a target figure by using an initial training set and adopting a semi-supervised semantic segmentation model;
the forgery detection and identification part performs dot multiplication on the mask data of the target person segmented according to the semantics and the face picture corresponding to the mask data to acquire a specified picture attribute region, and further performs model construction on the acquired region attribute, specifically:
for an original face picture z of each target figure, combining a picture mask a which is respectively divided by five sense organs in a mask data set with a manually selected interested five sense organ region vector V to obtain a facial five sense organ interested region, and then performing point multiplication on the facial five sense organ interested region and a corresponding original face picture to generate a condition tensor T of the required facial interested region;
performing point multiplication on an input picture z and a condition tensor T corresponding to the picture z, and inputting a point multiplication processing result p (z) into a generated countermeasure network to perform gesture-independent identification processing; the formula of dot product processing of the condition tensor T of the required face interesting region and the original image z is as follows:
p(z)＝z·T＝z·a·V
inputting p (z) and a given gesture into a generator G; generating a corresponding false picture by a generator G by using a given posture, judging the posture and the identity of the generated false picture by using a discriminator D, and continuously performing countermeasure training until the discriminator D judges that the false picture generated by the generator G is in a critical state with the same identity as the original input picture, so as to obtain a face picture irrelevant to the posture;
after the posture-independent recognition processing, inputting each face segmentation attribute region of the face picture x subjected to posture change into a single convolutional neural network for classification processing, and constructing a new CNN two-classification classifier; the image features are learned through a convolutional network, the output dimensionality is reduced through a pooling layer, the depth features are fused through a full-link layer, a classification result is finally formed and output, and the purpose of identifying whether the input image is a positive sample or a negative sample is achieved.
2. The method for detecting and identifying visual sense forgery of specific character based on semantic segmentation as claimed in claim 1, wherein: the semantic segmentation method comprises the following steps:
manually labeling N pictures of randomly extracted frames in a selected face image video, wherein the region of the 11-position part needs to be labeled in the manual labeling process; processing the json format file generated by labeling and the labeled picture original image together to obtain mask data sets of different face type labels;
selecting semantic segmentation network Deeplabv3+ training, inputting N mask data set pictures generated by manual labeling, and performing machine automatic labeling on the remaining M pictures which are not labeled in the face image through a deep learning model to realize labeling of semi-supervised machine learning, specifically:
two semantic segmentation networks P with the same structure and different weight initial values are constructed 1 And P 2 ：
P 1 ＝f(X；θ 1 )
P 2 ＝f(X；θ 2 )
Wherein X represents a pair of NThe marked picture is subjected to data enhancement to obtain an input picture; theta.theta. 1 And theta 2 Respectively represent P 1 And P 2 The weights of the two networks; y represents a pseudo label obtained by two semantic segmentation networks; for two split networks, obtaining corresponding one-hot label Y through argmax operation 1 And Y 2 (ii) a Then using the two pseudo labels as supervision signals and using Y 2 As P 1 Supervision of (A), Y 1 As P 2 And using cross entropy loss function to constrain; and finally, combining the M marked pictures generated by using the semantic segmentation network machine with the N originally and manually marked pictures to form a mask data set of the target person.
3. The method for detecting and identifying visual forgery of specific character based on semantic segmentation as claimed in claim 1, wherein: in the design of a classifier loss function, a binary cross entropy BCE loss function is used for measuring the classification loss, L Y Is defined as the following formula:
L Y (x,y)＝BCE(p,y)＝-(y*log(p))+(1-y)*log(1-p)
wherein x represents an input image, p is the predicted classification output of the classifier, y belongs to {0,1} and is a true and false label, and a sigmoid activation function is adopted in the two classification tasks to process the output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211188905.7A CN115482595B (en) | 2022-09-27 | 2022-09-27 | Specific character visual sense counterfeiting detection and identification method based on semantic segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211188905.7A CN115482595B (en) | 2022-09-27 | 2022-09-27 | Specific character visual sense counterfeiting detection and identification method based on semantic segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115482595A CN115482595A (en) | 2022-12-16 |
| CN115482595B true CN115482595B (en) | 2023-04-07 |
Family
ID=84393686
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211188905.7A Active CN115482595B (en) | 2022-09-27 | 2022-09-27 | Specific character visual sense counterfeiting detection and identification method based on semantic segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115482595B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116957919B (en) * | 2023-07-12 | 2024-07-16 | 珠海凌烟阁芯片科技有限公司 | RGBD image-based 3D human body model generation method and system |
| CN118379608B (en) * | 2024-06-26 | 2024-10-18 | 浙江大学 | High-robustness deep forgery detection method based on self-adaptive learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111985303A (en) * | 2020-07-01 | 2020-11-24 | 江西拓世智能科技有限公司 | Human face recognition and human eye light spot living body detection device and method |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10740596B2 (en) * | 2016-11-08 | 2020-08-11 | Nec Corporation | Video security system using a Siamese reconstruction convolutional neural network for pose-invariant face recognition |
| US11734955B2 (en) * | 2017-09-18 | 2023-08-22 | Board Of Trustees Of Michigan State University | Disentangled representation learning generative adversarial network for pose-invariant face recognition |
| CN113554045B (en) * | 2020-04-23 | 2024-04-09 | 国家广播电视总局广播电视科学研究院 | Data set manufacturing method, device, equipment and storage medium |
| CN113537027B (en) * | 2021-07-09 | 2023-09-01 | 中国科学院计算技术研究所 | Face depth counterfeiting detection method and system based on face division |
| CN113887573A (en) * | 2021-09-10 | 2022-01-04 | 合肥高维数据技术有限公司 | Human face forgery detection method based on visual converter |
| CN113989713B (en) * | 2021-10-28 | 2023-05-12 | 杭州中科睿鉴科技有限公司 | Depth forgery detection method based on video frame sequence prediction |
-
2022
- 2022-09-27 CN CN202211188905.7A patent/CN115482595B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111985303A (en) * | 2020-07-01 | 2020-11-24 | 江西拓世智能科技有限公司 | Human face recognition and human eye light spot living body detection device and method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115482595A (en) | 2022-12-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109583342B (en) | Human face living body detection method based on transfer learning | |
| Yang et al. | Detecting fake images by identifying potential texture difference | |
| CN115482595B (en) | Specific character visual sense counterfeiting detection and identification method based on semantic segmentation | |
| CN108537743A (en) | A kind of face-image Enhancement Method based on generation confrontation network | |
| Zheng et al. | Attention-based spatial-temporal multi-scale network for face anti-spoofing | |
| Liu et al. | Micro-expression recognition using advanced genetic algorithm | |
| de Souza et al. | On the learning of deep local features for robust face spoofing detection | |
| Roy et al. | 3D CNN architectures and attention mechanisms for deepfake detection | |
| Hebbale et al. | Real time COVID-19 facemask detection using deep learning | |
| Chen et al. | A pornographic images recognition model based on deep one-class classification with visual attention mechanism | |
| CN116563957B (en) | Face fake video detection method based on Fourier domain adaptation | |
| CN115240280A (en) | Construction method of human face living body detection classification model, detection classification method and device | |
| Mehra | Deepfake detection using capsule networks with long short-term memory networks | |
| CN112434647A (en) | Human face living body detection method | |
| Sharma et al. | Deepfakes Classification of Faces Using Convolutional Neural Networks. | |
| Peng et al. | Presentation attack detection based on two-stream vision transformers with self-attention fusion | |
| Qiao et al. | Fully unsupervised deepfake video detection via enhanced contrastive learning | |
| Kumar et al. | Mediapipe and cnns for real-time asl gesture recognition | |
| CN117935375A (en) | Anti-fake method for handwriting electronic signature | |
| Bikku et al. | Deep Residual Learning for Unmasking DeepFake | |
| CN111191549A (en) | Two-stage face anti-counterfeiting detection method | |
| Kanwal et al. | Exposing ai generated deepfake images using siamese network with triplet loss | |
| Saif et al. | Aggressive action estimation: a comprehensive review on neural network based human segmentation and action recognition | |
| CN114519897A (en) | Human face in-vivo detection method based on color space fusion and recurrent neural network | |
| CN112183357A (en) | Deep learning-based multi-scale in-vivo detection method and system |
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 |