CN115410260B - Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium - Google Patents
Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115410260B CN115410260B CN202211076090.3A CN202211076090A CN115410260B CN 115410260 B CN115410260 B CN 115410260B CN 202211076090 A CN202211076090 A CN 202211076090A CN 115410260 B CN115410260 B CN 115410260B
- Authority
- CN
- China
- Prior art keywords
- false
- true
- image
- face
- enhancement
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000012545 processing Methods 0.000 claims abstract description 18
- 241000282414 Homo sapiens Species 0.000 claims description 26
- 238000004891 communication Methods 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 12
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 10
- 239000000126 substance Substances 0.000 claims description 8
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 2
- 238000001514 detection method Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 7
- 238000010422 painting Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000009877 rendering Methods 0.000 description 3
- 108010068977 Golgi membrane glycoproteins Proteins 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000004397 blinking Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 210000004709 eyebrow Anatomy 0.000 description 1
- 210000000887 face Anatomy 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011425 standardization method Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
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
- G06V40/10—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- 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
- G06V40/40—Spoof detection, e.g. liveness detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present disclosure relates to a face authentication method and apparatus, an electronic device and a storage medium, wherein the method comprises: drawing a face envelope line on a suspected face-changing image, wherein the suspected face-changing image is obtained by using a trained face discrimination model for identification; taking a region in the face envelope range on the suspected face-changing image as a target region, and performing enhancement processing on the image of the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image; and comparing the more true enhancement image and the more false enhancement image with the suspected face-changed image to display the more true enhancement image and the more false enhancement image as an identification evidence of the suspected face-changed image, and comparing the more true enhancement image and the more false enhancement image of the suspected face-changed image with the image original image of the target area to highlight a fake trace of the suspected face-changed image as the identification evidence of the face identification model.
Description
Technical Field
The present disclosure relates to the field of image processing technologies, and in particular, to a method and an apparatus for face authentication and evidence collection, an electronic device, and a storage medium.
Background
The deep forgery refers to false human face portrait audio and video contents, such as face changing, expression control, mouth shape control, voice conversion and the like, manufactured based on a deep learning technology. In recent years, high-quality deep synthesis technology is rapidly developed, security threats brought by abuse of the technology are increasingly highlighted, and destructive influences may be brought to personal reputation, network information credibility, social stability and national security.
In the media forensics field, the detection and defense technical research on deep forged images and videos is very concerned, some methods provide effective detection schemes by using clues that forged faces lack blinking actions, lack pulse signals, inconsistent eye reflection directions and the like, and the clue-based methods have good interpretability, but have more defects and limitations in generality and detection efficiency. More researchers focus on the research of the false distinguishing model based on deep learning, and an effective deep false distinguishing model is trained on a large-scale true and false face image data set by adopting a data driving thought. The current famous deep fake face data set is widely used for training a deep fake identification model. In terms of model methods, researchers put forward many detection models with stronger detection performance and better generalization in the aspects of data augmentation, attention mechanism, detection and positioning multi-task learning, model integration and the like. The current deep false distinguishing model has made great progress in detection accuracy and starts to be widely deployed in practical application. However, due to the characteristics of the black box of the deep learning model and the imperceptibility of the forged trace, the detection result and basis of the counterfeit identification model are often difficult to be understood by human beings. This causes great troubles for application scenarios where interpretability is important, such as judicial law, and rumor, where the classification basis of the model is strongly demanded by users, and the lack of interpretability severely limits the deep application of the deep false identification model.
Can explain artificial intelligence, is the research hotspot of artificial intelligence academia. Most of the research on model interpretability in the current computer vision field is focused on application scenes such as object classification and recognition, and interpretable methods are roughly divided into two types, namely image area attribution and feature visualization. The region attribution method calculates and displays the contribution degree of each region of the image to the final classification result, and can generally highlight the local region of the image which is most important to the classification result. The characteristic visualization method obtains an image mode capable of maximally activating a specific neuron in the network through optimization so as to visually display what input mode each neuron in the network learns, and various standardization methods are proposed in the current research so as to improve the naturalness of a visualized image.
Although interpretability has been studied in the field of semantic object recognition, there has been little research in deep forgery detection. The method adopts the existing interpretable artificial intelligence method to carry out experimental verification on a deep counterfeiting detection model, adopts a plurality of area attribution class interpretation methods, and highlights which areas in a face image play a role in the classification result of the model. Although the methods can show the important regions, the human face regions are generally highlighted for true and false images, and the human vision is still insensitive to forged traces in the highlighted regions, so that the judgment basis of the model is difficult to effectively explain. For example, existing methods for interpreting and analyzing depth false distinguishing results focus on the positioning of important areas, and these methods can only give an explanation of which parts in a face image may have false marks, while for a face-changing image with higher quality, the human eyes usually cannot understand what abnormal marks exist in these areas.
Therefore, a method and an apparatus for face authentication and evidence collection, an electronic device and a storage medium are needed.
Disclosure of Invention
In order to solve the technical problem or at least partially solve the technical problem, embodiments of the present disclosure provide a method and an apparatus for face counterfeit detection and evidence obtaining, an electronic device, and a storage medium.
In a first aspect, an embodiment of the present disclosure provides a face counterfeit identification and forensics method, including:
drawing a face envelope line on a suspected face-changing image, wherein the suspected face-changing image is obtained by using a trained face discrimination model for identification;
taking a region in the face envelope range on the suspected face-changing image as a target region, and performing enhancement processing on the image of the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image;
and comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changing image to serve as the identification evidence of the suspected face-changing image.
In a possible implementation, the enhancing the image of the target area to obtain a truer enhanced image and a truer enhanced image of the suspected face-changed image respectively includes:
and adjusting at least one of the color parameter and the pixel value of the target area image, and respectively obtaining a more true enhancement image and a more false enhancement image corresponding to the target area image according to the adjusted at least one of the color parameter and the pixel value.
In a possible implementation manner, the adjusting the color parameter of the target area image to obtain a truer enhancement map and a truer enhancement map corresponding to the target area image according to the adjusted color parameter includes:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image being true or false to the color parameter according to the chain rule back-transmission gradient information of the face counterfeit identification model;
thirdly, according to the derivative of the suspected face-changing image to the color parameter, the color parameter is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false color parameter after preliminary adjustment is obtained;
step four, generating a color more true or more false enhancement map after the preliminary adjustment according to the color more true or more false parameter after the preliminary adjustment, inputting the color more true or more false enhancement map after the preliminary adjustment into the trained face discrimination model, acquiring the probability that the color more true or more false enhancement map after the preliminary adjustment is true or false, and calculating the difference value between the probability that the color more true or more false enhancement map after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the color more true or false enhancement map after the initial adjustment as a more true or false enhancement map corresponding to the target area image;
and when the difference is not within the preset threshold range, returning to the step two, and executing the step two to determine the derivative of the probability that the preliminarily adjusted color truer or more false enhancement image is true or false to the color parameter according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference between the probabilities that the color truer or more false enhancement image respectively corresponding to the last two times in the multiple adjustments is true or false is within the preset threshold range, wherein the color truer or more false enhancement image after the multiple adjustments is used as the truer or more false enhancement image corresponding to the target area image.
In one possible embodiment, a preliminary adjusted color truer or truer enhancement map is generated from the preliminary adjusted truer or truer color parameters by the following expression:
wherein the content of the first and second substances,
boost the map for a more true or false color after the preliminary adjustment>
For a suspected face-changed image, be>
Is a mask image which is suspected of changing face>
The more true or false color parameter after the initial adjustment. />
In a possible embodiment, the color parameter is initially adjusted in a direction in which the suspected face-changed image is more true or more false according to a derivative of the suspected face-changed image on the color parameter according to the probability that the suspected face-changed image is true or false, so as to obtain an initially adjusted more true or false color parameter by the following expression:
wherein the content of the first and second substances,
for a more true or more false color parameter after the preliminary adjustment, a determination is made whether the color value is greater than or equal to>
For the pre-adjusted color parameter, is adjusted>
For a single step adjustment of the step size, <' >>
Is the derivative of the probability of a suspected face-changed image being true or false with respect to the color parameter.
In a possible implementation manner, the adjusting the pixel value of the target area image and obtaining a truer enhancement map and a truer enhancement map corresponding to the target area image according to the adjusted pixel value includes:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image as true or false to the pixel value according to the chain rule reverse transmission gradient information of the face counterfeit identification model;
thirdly, according to the derivative of the suspected face-changing image to the pixel value, the pixel value is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false pixel value after preliminary adjustment is obtained;
step four, generating a pixel more true or more false enhancement image after the preliminary adjustment according to the pixel value after the preliminary adjustment, inputting the pixel more true or more false enhancement image after the preliminary adjustment into the trained face authentication model, acquiring the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the pixel more true or more false enhancement map after the initial adjustment as a more true or more false enhancement map corresponding to the target area image;
and when the difference value is not within the preset threshold value range, returning to the step two, and executing the step two to determine the derivative of the probability that the pixel more true or more false enhancement image after the initial adjustment is true or false to the pixel value according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference value between the probabilities that the pixel more true or more false enhancement image respectively corresponding to the last two times of adjustment is true or false in the preset threshold value range, wherein the pixel more true or more false enhancement image after the multiple times of adjustment is used as the more true or more false enhancement image corresponding to the target area image.
In one possible embodiment, a preliminary adjusted pixel truer or falser enhancement map is generated from the preliminary adjusted truer or falser pixel values by the following expression:
wherein the content of the first and second substances,
based on the preliminary adjusted pixel's more true or more false enhancement map, the image is asserted>
For a suspected face-changed image, be>
Is a mask image which is suspected of changing face>
To a more true or false pixel value after the initial adjustment.
In a possible embodiment, the pixel values are preliminarily adjusted towards a direction in which the suspected face-changed image is more true or false according to a derivative of the probability that the suspected face-changed image is true or false to the pixel values, so as to obtain preliminarily adjusted more true or false pixel values, by the following expression:
wherein, the first and the second end of the pipe are connected with each other,
for more true or false pixel values after the preliminary adjustment>
For the pixel value before adjustment, < >>
For a single step adjustment of the step size, <' >>
For the derivative of the probability of a suspected face-changed image being true or false with respect to the pixel value, <' > or>
For optimizing the number of iteration steps, ->
Is the amplitude of the disturbance.
In a second aspect, an embodiment of the present disclosure provides a face authentication device, including:
the drawing module is used for drawing a face envelope curve on a suspected face-changing image, wherein the suspected face-changing image is obtained by utilizing a trained face identification model for identification;
the processing module is used for taking a region in the face envelope range on the suspected face-changing image as a target region, and performing enhancement processing on the image of the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image;
and the display module is used for comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changing image to serve as the identification evidence of the suspected face-changing image.
In a third aspect, an embodiment of the present disclosure provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the communication bus;
a memory for storing a computer program;
and the processor is used for realizing the human face false distinguishing evidence obtaining method when executing the program stored in the memory.
In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the above-mentioned face counterfeit identification and forensics method.
Compared with the prior art, the technical scheme provided by the embodiment of the disclosure at least has part or all of the following advantages:
according to the face counterfeit identification evidence obtaining method, a face envelope line is drawn on a suspected face-changing image, wherein the suspected face-changing image is obtained by identification through a trained face counterfeit identification model; taking a region in the range of an envelope line of the face on the suspected face-changed image as a target region, and respectively carrying out enhancement processing on the image of the target region to obtain a more true enhancement image and a more false enhancement image of the suspected face-changed image; and comparing the more true enhancement image and the more false enhancement image with the suspected face-changed image to display the more true enhancement image and the more false enhancement image as an identification evidence of the suspected face-changed image, and comparing the more true enhancement image and the more false enhancement image of the suspected face-changed image with the image original image of the target area to highlight a fake trace of the suspected face-changed image as the identification evidence of the face identification model.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the related art will be briefly described below, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic flow chart illustrating a human face authentication method according to an embodiment of the disclosure;
fig. 2 schematically shows a flow chart of an actual application of the face counterfeit identification and forensics method according to the embodiment of the disclosure;
FIG. 3 schematically illustrates a contrast display of a more color true enhancement map, a suspected face-changed image, a more color false enhancement map, according to an embodiment of the disclosure;
FIG. 4 schematically illustrates a comparative display of a pixel-true enhancement map, a suspected face-changed image, a pixel-false enhancement map, according to an embodiment of the disclosure;
FIG. 5 is a block diagram schematically illustrating the structure of a human face counterfeit evidence collection apparatus according to an embodiment of the present disclosure; and
fig. 6 schematically shows a block diagram of an electronic device according to an embodiment of the present disclosure.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions of the embodiments of the present disclosure will be described clearly and completely with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are some embodiments of the present disclosure, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
Referring to fig. 1, an embodiment of the present disclosure provides a face authentication method, including:
s11, drawing a face envelope curve on a suspected face-changed image, wherein the suspected face-changed image is obtained by using a trained face identification model for identification;
s12, taking a region in the face envelope range on the suspected face-changing image as a target region, and performing enhancement processing on the image of the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image;
and S13, comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changing image to serve as an identification evidence of the suspected face-changing image.
In this embodiment, in step S12, the enhancing the image of the target area to obtain a truer enhanced image and a truer enhanced image of the suspected face-changed image respectively includes:
and adjusting at least one of the color parameter and the pixel value of the target area image, and respectively obtaining a more true enhancement image and a more false enhancement image corresponding to the target area image according to the adjusted at least one of the color parameter and the pixel value.
In this embodiment, the adjusting the color parameter of the target area image and obtaining a truer enhancement map and a truer enhancement map corresponding to the target area image according to the adjusted color parameter includes:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image being true or false to the color parameter according to the chain rule back-transmitted gradient information of the face false-distinguishing model;
thirdly, according to the derivative of the suspected face-changing image to the color parameter, the color parameter is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false color parameter after preliminary adjustment is obtained;
step four, generating a color more true or more false enhancement image after the preliminary adjustment according to the color more true or more false parameter after the preliminary adjustment, inputting the color more true or more false enhancement image after the preliminary adjustment into the trained face discrimination model, acquiring the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the color more true or false enhancement map after the initial adjustment as a more true or false enhancement map corresponding to the target area image;
and when the difference is not within the preset threshold range, returning to the step two, and executing the step two to determine the derivative of the probability that the preliminarily adjusted color truer or more false enhancement image is true or false to the color parameter according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference between the probabilities that the color truer or more false enhancement image respectively corresponding to the last two times in the multiple adjustments is true or false is within the preset threshold range, wherein the color truer or more false enhancement image after the multiple adjustments is used as the truer or more false enhancement image corresponding to the target area image.
In this embodiment, the color truer or false-more enhancement map after the preliminary adjustment is generated according to the color truer or false-more parameter after the preliminary adjustment by the following expression:
wherein the content of the first and second substances,
boost the map for a more true or false color after the preliminary adjustment>
For a suspected face-changed image, be>
Is a mask image which is suspected of changing face>
The more true or false color parameter after the initial adjustment.
In practical applications, referring to fig. 2, the calculation of the mask image and the more true enhancement map and the more false enhancement map of the suspected face-changed image includes:
suspected face-changing image detected by trained face false-distinguishing model
Carrying out automatic positioning calculation on 68 face key points, sequentially connecting eyebrow key points and face contour key points to form a face envelope line, filling pure white in the region inside the face envelope line, and filling pure black in the region outside the face envelope line to obtain a mask image of a suspected face-changing image/>
。
Six-dimensional color parameters for mask image
Adjusting, wherein each dimension color parameter represents a contrast adjustment parameter and a tone adjustment parameter of RGB three channels, and the adjustment target is to make the human face counterfeit distinguishing model respectively judge the enhanced image after adjustment to be more true and more false to obtain two color space counterfactual enhanced images, namely more true enhanced image/real image>
And a more false enhancement map>
。
The color space counterfactual enhanced image is constructed by expression
The region outside the face envelope of the original image is reserved, and only the RGB channels of the image in the face envelope are respectively subjected to contrast adjustment and tone adjustment, so that a more true and false enhanced image in a color space can be obtained through adjustment.
In this embodiment, the color parameter is initially adjusted in a direction in which the suspected face-changed image is more true or more false according to a derivative of the suspected face-changed image to the color parameter according to the probability that the suspected face-changed image is true or false, so as to obtain an initially adjusted more true or more false color parameter, according to the following expression:
wherein, the first and the second end of the pipe are connected with each other,
for a more true or false color parameter after the preliminary adjustment>
For the pre-adjusted color parameter, is adjusted>
For a single step adjustment of the step size, <' >>
Is the derivative of the probability of a suspected face-changed image being true or false with respect to the color parameter.
In practical applications, the adjustment process of the color space counterfactual enhanced image may be manual adjustment, such as manual color adjustment in image editing software, or may be an automatic optimization method. Inputting a suspected face-changing image into a trained face identification model in automatic optimization to obtain the probability that the image is judged to be true or false
Then reversely transmitting the gradient information according to a deep network chain rule to obtain the probability to the color parameter->
According to the derivative, the color parameter is subjected to iterative optimization towards a more true or false direction to obtain the optimal color parameter
Wherein the tag variable>
Real or fake. And finally, applying the optimized optimal color parameters to the image to generate a truer or fake counterfactual enhanced image. In more detail, the algorithm for automatically optimizing the color space enhanced image is described as follows:
input is suspected face-changed image
Mask image of the suspected face-changed image->
Fixing the trained face discrimination model>
Single step optimization step->
Tolerance of stopping conditions>
。/>
output more true enhancement map
Even more false enhancement map>
。
1 for tag variable
∈ [real, fake] do
2. Initializing color parameters
；
3. Initialization
;
4 repeat
5 
;
6. Using current parameters and construct expressions
Evaluating an enhanced image pick>
;
7. The calculation model judges that the current enhanced image is a label
Has a probability of->
;
8. Calculating partial derivatives of current probabilities to color parameters
;
10. Updating color parameters
;
12 The until satisfies the stop condition in 5 continuous steps
;
13 
;
14 end
The method for manually toning in the image editing software has a more flexible form, for example, a suspected face-changing image is opened in the Photoshop, GIMP and other software, a target area image is obtained through mouse operation, then some adjustments are performed on the target area image by using software tools such as color curves, tone adjustment, contrast adjustment and the like, so that the adjusted image has more (or less) color matching reality, and the result obtained by manually toning is a more true enhancement image and a more false enhancement image in a color space.
In this embodiment, the adjusting the pixel value of the target area image and obtaining a truer enhancement map and a truer enhancement map corresponding to the target area image according to the adjusted pixel value includes:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changed image being true or false to the pixel value according to the chain rule back-transmitted gradient information of the face false-identification model;
thirdly, according to the derivative of the suspected face-changing image to the pixel value, the pixel value is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false pixel value after preliminary adjustment is obtained;
step four, generating a pixel more true or more false enhancement image after the preliminary adjustment according to the pixel value after the preliminary adjustment, inputting the pixel more true or more false enhancement image after the preliminary adjustment into the trained face authentication model, acquiring the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the pixel more true or more false enhancement map after the initial adjustment as a more true or more false enhancement map corresponding to the target area image;
and when the difference value is not within the preset threshold value range, returning to the step two, and executing the step two to determine the derivative of the probability that the pixel more true or more false enhancement image after the initial adjustment is true or false to the pixel value according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference value between the probabilities that the pixel more true or more false enhancement image respectively corresponding to the last two times of adjustment is true or false in the preset threshold value range, wherein the pixel more true or more false enhancement image after the multiple times of adjustment is used as the more true or more false enhancement image corresponding to the target area image.
In this embodiment, a more true or more false enhancement map of the preliminarily adjusted pixel is generated from the more true or more false pixel value after the preliminary adjustment by the following expression:
wherein the content of the first and second substances,
based on the preliminary adjusted pixel's more true or more false enhancement map, the image is asserted>
For a suspected face-changed image, be>
Is a mask image which is suspected of changing face>
To a more true or false pixel value after the initial adjustment.
In practical applications, referring to FIG. 2, the image pixel values for the target area
Adjusting the image to make the human face discrimination model judge the enhanced image to be more true and more false respectively to obtain two pixel space counterfactual enhanced images, namely more true enhanced image based on whether or not the image is true>
And a more false enhancement map>
；
The pixel space counterfactual enhanced image is constructed in the following mode:
namely, the pixel value of the image in the face envelope curve is subjected to disturbance adjustment, and an enhanced image with more true and false pixel space can be obtained through adjustment.
In this embodiment, the pixel value is initially adjusted in a direction in which the suspected face-changed image is more true or more false according to a derivative of the suspected face-changed image with respect to the pixel value according to the probability that the suspected face-changed image is true or false, so as to obtain an initially adjusted more true or false pixel value, according to the following expression:
wherein the content of the first and second substances,
for a more true or more false pixel value after the preliminary adjustment, a value is selected>
For pre-adjusted pixel values>
For a single step adjustment of the step size, <' >>
For the derivative of the probability of a suspected face-changed image being true or false with respect to the pixel value, <' > or>
For optimizing the number of iteration steps, ->
Is the amplitude of the disturbance.
In practical application, the optimization process of the pixel space counterfactual enhanced image can also be a manual adjustment method, such as manual painting in image editing software, or can also be manual paintingIs an automatic optimization method. Inputting a suspected face-changing image into a trained face identification model in automatic optimization to obtain the probability that the image is judged to be true or false
Then, the gradient information is reversely transmitted according to the chain rule of the depth network to obtain the probability to the pixel value->
According to the derivative, the pixel value is iteratively optimized in the direction of more true or false in order to obtain an optimum pixel value->
Wherein the tag variable>
To real or fake, the optimized optimal pixel values are finally applied to the suspect face image to generate a more true or false counterintuitive enhanced image->
And &>
. In addition, the method for automatically optimizing the pixel space enhanced image can add random white noise on the original image for multiple times, carry out multiple independent optimization and finally average, so that a more stable enhancement effect can be obtained. The algorithm is described as follows:
input is a suspected face-changing imageIMask image of suspected face-changed imageMFixing the trained face discrimination modelfOptimizing the number of iteration stepsnAmplitude of disturbance
Number of random noise initializationsK。
output more true enhancement graph
Even more false enhancement map>
。/>
1 for tag variable
∈ [real, fake] do
2 for k = 1:K do
3. For the original imageIAdding random Gaussian noise;
4. initializing a disturbance image
;
5 for i = 1:n do
6. Using the current parameters and
evaluating an enhanced image pick>
;
7. The calculation model judges that the current enhanced image is a label
Probability of (2)
;
8. Calculating partial derivative of current probability to disturbed image
;
9. Updating an optimization step size
;
10. Updating a perturbed image
;
11 end
12 
;
13 end
14 
;
15 end
The method for manually painting in the image editing software has a more flexible form, for example, a suspected face-changing image is opened in the Photoshop, GIMP and other software, an image of a target area is obtained through mouse operation, and then some painting or painting adjustment is carried out on suspected pixel forged traces (such as unclear tooth particle lines, spliced edges, local artifacts and the like) of the face by using tools such as a painting brush, liquefaction, blurring and the like, so that the forged traces are more (or less) obvious, and the results obtained by manually painting are a pixel space truer enhanced image and a more false enhanced image.
Fig. 3 and 4 respectively show a color space interpretation result and a pixel space interpretation result, and the present disclosure performs enhancement display from two angles of inconsistent colors of the inner and outer faces and pixel traces of the face, where the middle image shown in fig. 3 and 4 is a suspicious face-changing image, the left and right sides are respectively a more true enhancement image and a more false enhancement image, and the lower part of each image is marked with a probability that the face-identifying model determines that the current image is false (the interval is 0~1). It can be seen that in fig. 3, the probability that the original suspected face-changed image is judged to be false by the false identification model is 0.91, which is much greater than the threshold of 0.5, so that the model considers that the image is a false image with a high probability, but when only the original image is observed, it is difficult for a human to see what false trace exists. The left and right images in fig. 3 show how the color change of the face region will appear to be more true and false than the original image, and it can be seen that the false distinguishing probability given by the false distinguishing model to the two enhanced images is 0.47 and 0.99, respectively. Meanwhile, the face color of the left image and the outer face are more harmonious, the inner face color and the outer face color of the right image are less consistent, and the fake trace of slight inconsistency of the face colors in the middle original image can be found by comparison. In fig. 4, after the face pixels of the original suspected face-changed image are enhanced, the right image highlights an artifact trace in the line frame region, the corresponding region of the left image inhibits the artifact, the false judgment probabilities of the false distinguishing model on the two true and false enhanced images are respectively 0.01 and 0.99, and the original image in the line frame region can be found to have a certain artifact trace by comparing the original image with the two enhanced images. Therefore, the examples in the two result graphs show the effect of the method of the present disclosure more intuitively.
In fig. 3 and 4, the color space enhanced image and the pixel space enhanced image are respectively displayed and output side by side with the original suspected face-changing image, and the judgment basis and the result of the face counterfeit identification model can be understood and analyzed by comparing and checking the difference of counterfeit traces before and after enhancement. In fact, when the enhanced image is displayed side by side with the original image, the three contrasting images in fig. 3 are, from left to right: more realistic image in color space
Original suspected face-changing imageIThe color space more false image->
(ii) a The three contrast images in fig. 4 are, from left to right: pixel space more true image->
Original suspected face-changing imageIPixel space more false image>
Simultaneously displaying the trained face authentication model at the bottom of each displayed image to judge the current image asProbability value of forgery. The difference of the forged traces of the color space and the pixel space before and after enhancement is contrasted and observed, and the judgment basis and the result of the counterfeit identification model can be understood and analyzed by combining the counterfeit judgment probability of the model on each image.
The human face counterfeit identification evidence obtaining method disclosed by the invention is based on an anti-fact image enhancement interpretation method to enhance suspicious counterfeit traces in the suspected face-changing image, and compares and views the counterfeit trace difference before and after enhancement to enable the suspicious counterfeit traces in the suspected face-changing image to be more easily observed by human eyes, so that the identification basis and result of the human face counterfeit identification model can be enhanced and explained by enhancing and displaying potential counterfeit traces.
The face counterfeit identification evidence obtaining method disclosed by the invention is based on the counter-fact optimization method, and at least one of inconsistent color traces and pixel traces in a suspicious forged image is enhanced to be used as the evidence of the face counterfeit identification model identification result, so that powerful support is provided for the face counterfeit identification model identification result.
Referring to fig. 5, an embodiment of the present disclosure further provides a face authentication evidence obtaining apparatus, including:
the drawing module 11 is configured to draw a face envelope curve on a suspected face-changed image, where the suspected face-changed image is obtained by using a trained face identification model for identification;
a processing module 12, configured to use a region in an envelope range of a face on the suspected face-changed image as a target region, perform enhancement processing on an image of the target region, and obtain a more true enhancement image and a more false enhancement image of the suspected face-changed image respectively;
and the display module 13 is used for comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changed image as the identification evidence of the suspected face-changed image.
The implementation process of the functions and actions of each unit in the above device is specifically described in the implementation process of the corresponding step in the above method, and is not described herein again.
For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the invention. One of ordinary skill in the art can understand and implement it without inventive effort.
In the second embodiment, any plurality of the rendering module 11, the processing module 12, and the display module 13 may be combined and implemented in one module, or any one of the modules may be divided into a plurality of modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of the other modules and implemented in one module. At least one of the rendering module 11, the processing module 12 and the display module 13 may be implemented at least partially as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented in hardware or firmware in any other reasonable manner of integrating or packaging a circuit, or in any one of three implementations of software, hardware and firmware, or in a suitable combination of any of them. Alternatively, at least one of the rendering module 11, the processing module 12 and the display module 13 may be at least partly implemented as a computer program module, which when executed may perform a corresponding function.
Referring to fig. 6, an electronic device provided by a fourth exemplary embodiment of the present disclosure includes a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, where the processor 1110, the communication interface 1120, and the memory 1130 complete communication with each other through the communication bus 1140;
a memory 1130 for storing computer programs;
the processor 1110 is configured to implement the following human face authentication evidence obtaining method when executing the program stored in the memory 1130:
drawing a face envelope line on a suspected face-changing image, wherein the suspected face-changing image is obtained by using a trained face discrimination model for identification;
taking a region in the face envelope range on the suspected face-changing image as a target region, and performing enhancement processing on the image of the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image;
and comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changing image to serve as the identification evidence of the suspected face-changing image.
The communication bus 1140 may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus 1140 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown, but this is not intended to represent only one bus or type of bus.
The communication interface 1120 is used for communication between the electronic device and other devices.
The Memory 1130 may include a Random Access Memory (RAM) and a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. Optionally, the memory 1130 may also be at least one memory device located remotely from the processor 1110.
The Processor 1110 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the Integrated Circuit may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component.
Embodiments of the present disclosure also provide a computer-readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the human face false-identification evidence-obtaining method.
The computer-readable storage medium may be contained in the apparatus/device described in the above embodiments; or may be separate and not incorporated into the device/apparatus. The computer-readable storage medium carries one or more programs which, when executed, implement the method for face authentication and forensics according to the embodiments of the present disclosure.
According to embodiments of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
The foregoing are merely exemplary embodiments of the present disclosure, which enable those skilled in the art to understand or practice the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A face false-distinguishing evidence-obtaining method is characterized by comprising the following steps:
drawing a face envelope line on a suspected face-changing image, wherein the suspected face-changing image is obtained by using a trained face discrimination model for identification;
taking a region in the range of the face envelope line on the suspected face-changing image as a target region, and performing enhancement processing on the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image;
comparing the more true enhancement image and the more false enhancement image with the suspected face-changed image to display the more true enhancement image and the more false enhancement image as the identification evidence of the suspected face-changed image,
wherein, the enhancement processing is carried out on the target area to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changing image, and the method comprises the following steps:
adjusting at least one of the color parameter and the pixel value of the target area, respectively obtaining a more true enhancement map and a more false enhancement map corresponding to the target area according to the adjusted at least one of the color parameter and the pixel value,
adjusting the color parameters of the target area, and obtaining a truer enhancement map and a fake enhancement map corresponding to the target area according to the adjusted color parameters, wherein the method comprises the following steps:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image being true or false to the color parameter according to the chain rule back-transmitted gradient information of the face false-distinguishing model;
thirdly, according to the derivative of the suspected face-changing image to the color parameter, the color parameter is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false color parameter after preliminary adjustment is obtained;
step four, generating a color more true or more false enhancement image after the preliminary adjustment according to the color more true or more false parameter after the preliminary adjustment, inputting the color more true or more false enhancement image after the preliminary adjustment into the trained face discrimination model, acquiring the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the color more true or false enhancement map after the initial adjustment as a more true or false enhancement map corresponding to the target area;
when the difference value is not in the preset threshold value range, returning to the step two to execute the step of determining the derivative of the probability that the color more true or more false enhancement image after the initial adjustment is true or false to the color parameter according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference value between the probabilities that the color more true or more false enhancement image respectively corresponding to the last two times in the multiple adjustments is in the preset threshold value range, and taking the color more true or more false enhancement image after the multiple adjustments as the more true or more false enhancement image corresponding to the target area,
adjusting the pixel value of the target area, and obtaining a truer enhancement map and a fake enhancement map corresponding to the target area according to the adjusted pixel value, wherein the method comprises the following steps:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image as true or false to the pixel value according to the chain rule reverse transmission gradient information of the face counterfeit identification model;
thirdly, performing primary adjustment on the pixel value in a direction of trueness or falseness of the suspected face-changed image according to a derivative of the suspected face-changed image on the pixel value to obtain a trueness or falseness pixel value after the primary adjustment;
step four, generating a pixel more true or more false enhancement image after the initial adjustment according to the pixel value after the initial adjustment, inputting the pixel more true or more false enhancement image after the initial adjustment into the trained face discrimination model, acquiring the probability that the pixel more true or more false enhancement image after the initial adjustment is true or false, and calculating the difference value between the probability that the pixel more true or more false enhancement image after the initial adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the pixel more true or more false enhancement map after the initial adjustment as a more true or more false enhancement map corresponding to the target area;
and when the difference is not within the preset threshold range, returning to the step two, and executing the step two to determine the derivative of the probability that the pixel more true or more false enhancement image after the initial adjustment is true or false to the pixel value according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference between the probabilities that the pixel more true or more false enhancement image respectively corresponding to the last two times in the multiple adjustments is true or false is within the preset threshold range, and taking the pixel more true or more false enhancement image after the multiple adjustments as the more true or more false enhancement image corresponding to the target area.
2. The method of claim 1, wherein the preliminary adjusted color truer or truer enhancement map is generated from the preliminary adjusted truer or truer color parameters by the following expression:
wherein the content of the first and second substances,
the color of the image is adjusted to be more true or false, I is a suspected face-changing image, M is a mask image of the suspected face-changing image, and alpha is r ，α g ，α b ，β r ，β g ，β b The more true or false color parameter after the initial adjustment.
3. The method according to claim 1, wherein the preliminary adjustment of the color parameter towards the direction of the suspected face-changed image being more true or more false is performed according to the derivative of the probability of the suspected face-changed image being true or false on the color parameter by the following expression to obtain a preliminary adjusted more true or false color parameter:
γ i ＝γ i-1 +η·d γ /||d γ ||
wherein, γ i For more true or false color parameters after preliminary adjustment, gamma i-1 For the color parameters before adjustment, η is the step size of the single step adjustment, d γ The derivative of the probability of a suspected face change image being true or false with respect to the color parameter.
4. The method of claim 1, wherein the preliminary adjusted pixel truer or truer enhancement map is generated from the preliminary adjusted truer or truer pixel values by the following expression:
wherein the content of the first and second substances,
the image is a more true or false enhancement image of the pixels after the initial adjustment, I is a suspected face-changing image, M is a mask image of the suspected face-changing image, and delta is a more true or false pixel value after the initial adjustment.
5. The method according to claim 1, wherein the preliminary adjustment of the pixel values in the direction of trueness or falseness of the suspected face-changed image is performed according to the derivative of the probability of true or false of the suspected face-changed image on the pixel values by the following expression to obtain the preliminarily adjusted trueness or falseness pixel values:
δ i ＝δ i-1 -η·d δ /||d δ ||
η＝2∈·(n-i-1)/n 2
wherein, delta i For more true or false pixel values after preliminary adjustment, δ i-1 For the pixel values before adjustment, η is the step size of the single step adjustment, d δ And (4) the derivative of the probability of the suspected face-changing image being true or false to the pixel value, n is the number of the optimization iteration steps, and epsilon is the disturbance amplitude.
6. A face false-distinguishing evidence-obtaining device is characterized by comprising:
the system comprises a drawing module, a face identification module and a face identification module, wherein the drawing module is used for drawing a face envelope curve on a suspected face-changing image, and the suspected face-changing image is obtained by utilizing a trained face identification model;
the processing module is used for taking a region in the envelope range of the face on the suspected face-changed image as a target region, and performing enhancement processing on the target region to respectively obtain a more true enhancement image and a more false enhancement image of the suspected face-changed image;
a display module for comparing and displaying the more true enhancement image and the more false enhancement image with the suspected face-changed image as the identification evidence of the suspected face-changed image,
wherein, the enhancing treatment is carried out on the target area to respectively obtain a more true enhanced image and a more false enhanced image of the suspected face-changing image, and the method comprises the following steps:
adjusting at least one of the color parameter and the pixel value of the target area, respectively obtaining a more true enhancement map and a more false enhancement map corresponding to the target area according to the adjusted at least one of the color parameter and the pixel value,
adjusting the color parameters of the target area, and obtaining a more true enhancement map and a more false enhancement map corresponding to the target area according to the adjusted color parameters, wherein the method comprises the following steps:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image being true or false to the color parameter according to the chain rule back-transmission gradient information of the face counterfeit identification model;
thirdly, according to the derivative of the suspected face-changing image to the color parameter, the color parameter is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false color parameter after preliminary adjustment is obtained;
step four, generating a color more true or more false enhancement image after the preliminary adjustment according to the color more true or more false parameter after the preliminary adjustment, inputting the color more true or more false enhancement image after the preliminary adjustment into the trained face discrimination model, acquiring the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the color more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the color more true or false enhancement map after the initial adjustment as a more true or false enhancement map corresponding to the target area;
when the difference value is not in the preset threshold value range, returning to the step two to execute the step of determining the derivative of the probability that the color more true or more false enhancement image after the initial adjustment is true or false to the color parameter according to the chain rule reverse transmission gradient information of the human face false distinguishing model until the difference value between the probabilities that the color more true or more false enhancement image respectively corresponding to the last two times in the multiple adjustments is in the preset threshold value range, and taking the color more true or more false enhancement image after the multiple adjustments as the more true or more false enhancement image corresponding to the target area,
adjusting the pixel value of the target area, and obtaining a truer enhancement map and a fake enhancement map corresponding to the target area according to the adjusted pixel value, wherein the method comprises the following steps:
acquiring the probability of identifying a suspected face-changing image as true or false by using a trained face identification model;
determining the derivative of the probability of the suspected face-changing image as true or false to the pixel value according to the chain rule reverse transmission gradient information of the face counterfeit identification model;
thirdly, according to the derivative of the suspected face-changing image to the pixel value, the pixel value is preliminarily adjusted towards the direction of the suspected face-changing image to be more true or more false, and the more true or more false pixel value after preliminary adjustment is obtained;
step four, generating a pixel more true or more false enhancement image after the preliminary adjustment according to the pixel value after the preliminary adjustment, inputting the pixel more true or more false enhancement image after the preliminary adjustment into the trained face authentication model, acquiring the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false, and calculating the difference value between the probability that the pixel more true or more false enhancement image after the preliminary adjustment is true or false and the probability that the suspected face-changed image is true or false:
when the difference value is within a preset threshold value range, taking the pixel more true or more false enhancement map after the initial adjustment as a more true or more false enhancement map corresponding to the target area;
and when the difference value is not in the preset threshold value range, returning to the step two to execute the step of reversely transmitting gradient information according to a chain rule of the human face discrimination model to determine the derivative of the probability that the pixel value is true or false of the pixel more true or false enhancement image after the initial adjustment until the difference value between the probabilities that the pixel more true or false enhancement image corresponding to the last two times of adjustment is true or false is in the preset threshold value range, and taking the pixel more true or false enhancement image after the multiple adjustments as the more true or false enhancement image corresponding to the target area.
7. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;
a memory for storing a computer program;
a processor for implementing the method for authenticating and evidence-obtaining human face according to any one of claims 1 to 5 when executing the program stored in the memory.
8. A computer-readable storage medium, on which a computer program is stored, the computer program, when being executed by a processor, implementing the method for human face authentication according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211076090.3A CN115410260B (en) | 2022-09-05 | 2022-09-05 | Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211076090.3A CN115410260B (en) | 2022-09-05 | 2022-09-05 | Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115410260A CN115410260A (en) | 2022-11-29 |
| CN115410260B true CN115410260B (en) | 2023-03-28 |
Family
ID=84164251
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211076090.3A Active CN115410260B (en) | 2022-09-05 | 2022-09-05 | Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115410260B (en) |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9075975B2 (en) * | 2012-02-21 | 2015-07-07 | Andrew Bud | Online pseudonym verification and identity validation |
| CN109558813A (en) * | 2018-11-14 | 2019-04-02 | 武汉大学 | A kind of AI depth based on pulse signal is changed face video evidence collecting method |
-
2022
- 2022-09-05 CN CN202211076090.3A patent/CN115410260B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115410260A (en) | 2022-11-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | Fake face detection via adaptive manipulation traces extraction network | |
| CN109587372B (en) | Invisible image steganography based on generation of countermeasure network | |
| Li et al. | Benchmarking single-image dehazing and beyond | |
| CN105917353A (en) | Feature extraction and matching and template update for biometric authentication | |
| WO2021137946A1 (en) | Forgery detection of face image | |
| CN110175646B (en) | Multi-channel confrontation sample detection method and device based on image transformation | |
| US20210390667A1 (en) | Model generation | |
| WO2020159437A1 (en) | Method and system for face liveness detection | |
| Roy et al. | 3D CNN architectures and attention mechanisms for deepfake detection | |
| CN110135446A (en) | Method for text detection and computer storage medium | |
| Mazumdar et al. | Universal image manipulation detection using deep siamese convolutional neural network | |
| Swami et al. | Candy: Conditional adversarial networks based fully end-to-end system for single image haze removal | |
| Rhee | Detection of spliced image forensics using texture analysis of median filter residual | |
| CN116453232A (en) | Face living body detection method, training method and device of face living body detection model | |
| Ahmed et al. | Evaluating the effectiveness of rationale-augmented convolutional neural networks for deepfake detection | |
| Nowroozi et al. | Detecting high-quality GAN-generated face images using neural networks | |
| CN116958782A (en) | Method and device for detecting weak and small targets by combining infrared and visible light characteristics | |
| Wang et al. | An eyes-based siamese neural network for the detection of gan-generated face images | |
| Dai et al. | A gated cross-domain collaborative network for underwater object detection | |
| CN115410260B (en) | Face counterfeit discrimination and evidence obtaining method and device, electronic equipment and storage medium | |
| CN112200075B (en) | Human face anti-counterfeiting method based on anomaly detection | |
| CN111275642B (en) | Low-illumination image enhancement method based on significant foreground content | |
| Wang et al. | Is underwater image enhancement all object detectors need? | |
| CN115936961B (en) | Steganalysis method, equipment and medium based on few-sample comparison learning network | |
| CN111191549A (en) | Two-stage face anti-counterfeiting detection method |
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 |