Machine learning method and device for image processing and training system

Abstract

One or more embodiments of the present specification relate to a machine learning method, apparatus and training system for image processing. The method is used for a training system comprising a detector and a forgery device, and comprises the following steps: updating the detector based on a real face sample set and a first forgery face sample set generated by a plurality of forgery parameters in a configuration pool of the forgery device; updating a forgery parameter in a configuration pool of the forgery device based on the updated detector; and updating the detector again based on the real face sample set and a second counterfeit face sample set generated by the counterfeit parameters in the updated configuration pool of the counterfeit device.

Description

Technical Field

[0001] The embodiments of the present specification generally relate to the field of computer technology, and more specifically to a machine learning method, device, and training system for image processing. Background Art

[0002] Deepfake is a combination of "deep learning" and "fake". It is a product of the development of artificial intelligence and refers to artificial intelligence technology that forges digital content through deep learning. It uses machine learning models to merge and superimpose images or videos onto source images or videos, uses neural network technology to learn large facial samples, and splices individual voices, facial expressions, and body movements into fake content to achieve a fake-to-real effect. It is called "PS technology in the field of audio and video" and its core feature is high simulation.

[0003] The most common illegal use of deep fake technology is to forge pornographic films for illegal profit, or to forge spoof videos, which constitutes a serious infringement on personal reputation and portrait rights. In addition, criminals can also exploit loopholes to hijack the mobile phone recognition camera, use deep fake technologies such as photo activation and expression manipulation to impersonate the owner of the phone, and then conduct transfer fraud on the owner's friends. Deep fake technology also poses a huge threat to the safety of public property. For example, hackers can use audio generation technology to impersonate senior executives to enhance their corporate email leakage attacks. What's more serious is that deep fake technology may also be used to create false news, thereby intensifying social conflicts and even posing a threat to national security. Summary of the invention

[0004] In view of this, one or more embodiments of the present specification provide a machine learning method, device and training system for image processing, which can use multiple forgery parameters in the configuration pool of the forger to generate a forged sample set based on a real face sample set and train a detector, and then update the forgery parameters of the forger based on the trained detector to generate more challenging forged samples to continue training the detector, thereby avoiding blind spots or over-reliance on specific tampering patterns of a certain data set, enriching the diversity of the forgery data space, enhancing sensitivity to forgery clues, and thereby improving the generalization ability to unknown attack types.

[0005] In a first aspect of the present specification, a machine learning method for image processing is provided. The method is used for a training system including a detector and a forger. The method includes performing the following actions at least once: updating the detector based on a real face sample set and a first forged face sample set generated by a plurality of forged parameters in a configuration pool of the forger; updating the forged parameters in the configuration pool of the forger based on the updated detector; and updating the detector again based on a real face sample set and a second forged face sample set generated by the forged parameters in the configuration pool of the updated forger.

[0006] In a second aspect of the present specification, a machine learning device for image processing is provided. The device includes a first detector update unit, which is configured to update the detector based on a real face sample set and a first forged face sample set generated by multiple forged parameters in a configuration pool of the forger. The device also includes a forger update unit, which is configured to update the forgery parameters in the configuration pool of the forger based on the updated detector. In addition, the device also includes a second detector update unit, which is configured to update the detector again based on the real face sample set and a second forged face sample set generated by the forged parameters in the configuration pool of the updated forger.

[0007] In a third aspect of the present specification, a computing device is provided, comprising: at least one processing unit; at least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the computing device to execute a method as described in the first aspect of the present specification.

[0008] In a fourth aspect of the present specification, a computer program product is provided, comprising machine executable instructions, which, when executed by a device, cause the device to perform the method according to the first aspect of the present specification.

[0009] In a fifth aspect of the present specification, a training system is provided. The training system includes a forger, the forger includes a configuration pool, the configuration pool includes a plurality of forgery parameters, and the forger is used to generate a forged face sample set based on a real face sample set and the forged parameters of the configuration pool. The training system also includes a detector, which is used to perform a binary classification operation on the input face sample and to perform a forged parameter reconstruction operation on the forged face sample in the input face sample.

[0010] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of this specification, nor are they intended to limit the scope of this disclosure. Other features of this specification will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The above and other purposes, features and advantages of the various embodiments of the present specification will become more easily understood through the following detailed description with reference to the accompanying drawings. In the accompanying drawings, various embodiments of the present specification will be described in an exemplary and non-limiting manner, wherein:

[0012] Figure 1 A schematic diagram of the structure of a training system according to some embodiments of the present specification is shown;

[0013] Figure 2 A flowchart of a machine learning method for image processing according to some embodiments of the present specification is shown;

[0014] Figure 3 A schematic diagram of the structure of a configuration pool of a forger according to some embodiments of the present specification is shown;

[0015] Figure 4 A specific structural schematic diagram of a training system according to some embodiments of this specification is shown;

[0016] Figure 5 A schematic diagram showing a training process of a training system according to some embodiments of the present specification;

[0017] Figure 6 A schematic block diagram showing a machine learning apparatus for image processing according to some embodiments of the present specification; and

[0018] Figure 7 A block diagram of an electronic device according to some embodiments of the present specification is shown.

[0019] Throughout the drawings, the same or similar reference numbers denote the same or similar elements. DETAILED DESCRIPTION

[0020] The embodiments of the present specification will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present specification are shown in the accompanying drawings, it should be understood that the present specification can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present specification. It should be understood that the drawings and embodiments of the present specification are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0021] In the description of the embodiments of this specification, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0022] As mentioned above, deep fakes, as a counterfeiting technology that manipulates and tampered with audio and video media data, have caused great social harm. However, the training method currently used by deep fake detection models based on convolutional neural networks is generally to generate a large amount of deep fake attack data offline first, and then train the model. Although this development framework that separates the attack data generation link and the defense model training link is very common, it has great limitations. First, the versatility and generalization of the model will be severely limited by the richness of the training data and the diversity of the tampering methods; second, when the model is actually applied, it has weak defense capabilities against tampering attack types that have not been seen during training.

[0023] To this end, a machine learning method for image processing is provided in an embodiment of the present specification. The method first generates a forged sample set based on a real face sample set using multiple forged parameters in the configuration pool of the forger and trains the detector, and then updates the forgery parameters of the forger based on the trained detector to generate more challenging forged samples to continue training the detector. In this way, blind spots or over-reliance on a specific tampering pattern of a certain data set can be avoided, the diversity of the forged data space can be enriched, the sensitivity to forgery clues can be enhanced, and the generalization ability of unknown attack types can be improved.

[0024] Figure 1 FIG. 1 is a schematic diagram showing a structure of a training system 100 according to some embodiments of the present specification. Figure 1 As shown, the training system 100 may include a real sample set 102, a forger 104, a forged sample set 106, and a detector 108. The real sample set 102 may include a plurality of real collected and untampered image samples. For example, the real sample set 102 may include a plurality of real taken face image samples.

[0025] like Figure 1 As shown, the forger 104 can tamper with the real image samples in the real sample set 102 based on the forgery parameters to obtain the forged sample set 106. The forgery parameters may include but are not limited to the tampering area, the tampering method, the fusion ratio, etc. For example, for a face image sample, the forger 104 can generate a forged face image sample by fusing specific facial regions of two face images. In addition, the forger 104 can also generate different forged face image samples by adjusting the fusion ratio or fusion method.

[0026] like Figure 1As shown, the real sample set 102 and the forged sample set 106 can be provided to the detector 108 for binary classification detection of real images and forged images. The detector 108 can be specifically implemented as a deep forgery detection model. When the forgery parameters of the forger 104 are fixed, the network parameters of the detector 108 can be trained to improve the accuracy of the binary classification detection. In some embodiments, the detector 108 can also perform a reconstruction operation on the forgery parameters of the forger 104 to enhance the sensitivity to forgery clues.

[0027] When the network parameters of the detector 108 are trained to a certain extent, the network parameters of the detector 108 can be fixed, and the forgery parameters of the forger 104 can be trained, so that the forger 104 can generate more challenging forged image samples based on the real sample set 102 to train the detector 108 again. Through continuous iteration, the richness of the training samples can be enhanced and the overfitting problem can be avoided.

[0028] It should be understood that the architecture and functions in the training system 100 are described for exemplary purposes only, and do not imply any limitation on the scope of the present specification. The embodiments of the present specification can also be applied to other training systems with different structures and / or functions.

[0029] The following will combine Figures 2 to 7 The process according to the embodiment of this specification is described in detail. For ease of understanding, the specific data mentioned in the following description are exemplary and are not intended to limit the scope of protection of the present disclosure. It should be understood that the embodiments described below may also include additional actions not shown and / or may omit the actions shown, and the scope of the present disclosure is not limited in this respect.

[0030] Figure 2 2 shows a schematic flow chart of a machine learning method 200 for image processing according to some embodiments of the present specification. The method 200 may be performed by, for example Figure 1 The training system 100 including the counterfeiter and the detector is implemented as shown. It should be understood that the method 200 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present disclosure is not limited in this respect.

[0031] like Figure 2 As shown, at block 210, method 200 may update the detector based on the real face sample set and the first forged face sample set generated by multiple forgery parameters in the configuration pool of the forger. In some embodiments, the detector may be implemented as a deep forgery detection model.

[0032] In some embodiments, the type of forged parameters in the configuration pool includes at least one of the following: forged object, forged area, fusion type, and fusion ratio. Using multiple different types of forged parameters in the configuration pool to tamper with real face samples can improve the quality of forged face samples compared to using random noise, thereby further improving the training efficiency of the detector.

[0033] In some embodiments, at the beginning of iterative training, the system 100 can randomly initialize the fake parameters in the configuration pool, and then generate a first fake face sample set based on the real face sample set and the initialized fake parameters. Then, the first fake face sample set can be used to preliminarily train the network parameters of the detector.

[0034] In some embodiments, the detector can be trained to perform the following operations: determine a first loss based on a binary classification result for an input face sample; and determine a second loss based on a forgery parameter reconstructed for a forged face sample in the binary classification result. By giving the detector an additional reconstruction task, the detector's sensitivity to forgery clues can be better enhanced.

[0035] In some embodiments, the system 100 can update the network parameters of the detector based on the first loss and the second loss when the fake parameters in the configuration pool are fixed. By optimizing the network parameters of the detector to minimize the sum of the first loss and the second loss, the detector parameters under the current iteration round can be trained.

[0036] In block 220, method 200 may update the forgery parameters in the configuration pool of the forger based on the updated detector. System 100 may update the forgery parameters of the configuration pool of the forger based on the first loss and the second loss while fixing the parameters of the detector. By optimizing the forgery parameters of the forger to maximize the sum of the first loss and the second loss, the forgery parameters under the current iteration round can be trained, thereby generating more challenging forged samples.

[0037] In block 230, method 200 can update the detector again based on the real face sample set and the second forged face sample set generated by the forged parameters in the configuration pool of the updated forger. The detector trained based on more challenging forged samples can better cope with unknown tampering types than the detector obtained using the fixed training set.

[0038] In this way, multiple forgery parameters in the forger's configuration pool can be used to generate a forged sample set based on the real face sample set and train the detector, and then the forgery parameters of the forger can be updated based on the trained detector to generate more challenging forged samples to continue training the detector. This can avoid blind spots or over-reliance on specific tampering patterns of a certain data set, enrich the diversity of the forgery data space, enhance sensitivity to forgery clues, and thereby improve the generalization ability to unknown attack types.

[0039] Figure 3 FIG. 3 shows a schematic diagram of a configuration pool 300 of a counterfeiter according to some embodiments of the present specification. Figure 3 As shown, the configuration pool 300 includes multiple types of forgery parameters, such as a forged object 302, a forged area 304, a fusion type 306, and a fusion ratio 308. Among them, the forged object 302 may include a forgery 302-2 based on a single image and a forgery 302-4 based on two images. The forgery 302-2 based on a single image can be implemented as fusing the features of a partial area of ​​the same image into other areas of the image, such as copying and flipping the left eye features of the target face image and fusing them into the right eye area of ​​the target face image. The forgery 302-4 based on two images can be implemented as fusing the features of a specific area of ​​one image into the same area of ​​another image, such as fusing the features of the mouth area of ​​a face image into the mouth area of ​​another face image.

[0040] The forged area 304 may include a random mask 304-2 and a selected area 304-4. The random mask 304-2 refers to a mask layer randomly generated on the image. This mask layer can be binary (i.e., black and white areas, black represents blocking, and white represents display), or it can be a grayscale image or a color image with different transparencies. By combining this random mask with the original image, such as multiplication, addition and other operations, the appearance of the original image is changed, thereby achieving the effect of image forgery. Due to its randomness, the concealment of image tampering can be increased. The selected area 304-4 can be implemented by selecting a specific facial feature area in, for example, a face image for tampering to obtain a forged image.

[0041] Fusion type 306 may include copy and paste 306-2, Alpha fusion 306-4 and Poisson fusion 306-6. Copy and paste 306-2 can copy a certain area (source area) in the image, and after pre-processing such as scaling, rotation, and transformation, paste it to another area (tampering area) in the image, thereby changing the original semantic information of the image. Alpha fusion 306-4 can add an alpha value to each pixel of the image in addition to red, green, and blue to store the transparency of the object. The two images are mixed by this transparency value, and the final color value of each pixel is determined by the color value of the corresponding pixel of the two images and their alpha values. Poisson fusion 306-6 can be based on the Poisson equation. While retaining the gradient information of the source image, the Poisson equation is solved according to the boundary conditions specified by the user, so that the brightness and color in the fusion area remain consistent, so as to achieve seamless fusion at the boundary. Due to the increase in the computational complexity of the above three different fusion methods, the forgery traces are more difficult to be detected.

[0042] The fusion ratio 308 may include a random value 308-2 between 0 and 1. If the value of the fusion ratio 308 is too high, after a large number of regions are copied and pasted, the overall layout and content of the image will appear extremely inconsistent, and the differences in light, shadow, color, etc. between the newly pasted portion and the original image will be magnified, seriously affecting the visual effect. If the value of the fusion ratio 308 is too low, although it is more difficult to detect, it cannot achieve the expected high-quality fusion effect, and the training will lose its meaning.

[0043] Figure 4 FIG. 4 shows a specific structural diagram of a training system 400 according to some embodiments of the present specification. Figure 4 As shown, the training system 400 mainly includes two parts: a forger 404 and a detector 408. The function of the forger 404 includes online generation of a forged sample set 406 based on the real sample set 402. Specifically, taking the face sample as an example, the face to be tampered, the area to be tampered, the fusion type and the fusion ratio when changing the face or pasting can be selected online based on the configuration pool. These forgery parameters are randomly selected when the first round of iterative training starts. After the attack and defense cycle, the forgery parameters can be learned in the direction that makes the overall constraint value 418 of the training system 400 larger.

[0044] The main functions of the detector 408 include a binary classification task 410, that is, determining whether the input image sample is real or forged. In order to prevent the detector 408 from falling into an idle state and to improve the sensitivity to forged details, the detector 408, in addition to completing the above-mentioned binary classification task, also needs to reconstruct the forged parameters (including image mask 412, fusion type 414, fusion ratio 416, etc.) of the image determined to be forged. Based on the binary classification task 410, the first loss 418-2 can be determined; based on the reconstruction task, the second loss 418-4 can be determined. The sum of the first loss 418-2 and the second loss 418-4 can be used as the overall constraint value 418 of the training system 400. In each iteration, the network parameters of the detector 408 can be learned in the direction that makes the overall constraint value 418 of the training system 400 smaller.

[0045] Figure 5 FIG. 5 is a schematic diagram showing a training process 500 of a training system according to some embodiments of the present specification. Figure 5 As shown, the first stage of the countermeasure attack and defense is the initialization stage, in which the forger 502 is initialized using random forged parameters.

[0046] Then, the system enters the second stage: the defense stage, fixing the forgery parameters of the forger and focusing on updating the detector 504. First, a batch of forged samples can be generated online using a randomly initialized forger, and used together with real image samples to train a round of detectors. At this time, the forgery parameters of the forger are fixed, and the network parameters of the detector are optimized to minimize the overall constraint value, while the network parameters are updated by gradient descent.

[0047] Then, the system enters the third stage: the attack stage, fixing the network parameters of the detector and focusing on updating the forger 506 to maximize the overall constraint value of the system. Since the operation process of forging and generating adversarial samples is not differentiable, the forged parameters of the configuration pool are also partially non-differentiable, and the back-propagation gradient flow will be disconnected, it is necessary to use a gradient approximation algorithm to approximate and optimize the gradient.

[0048] In some embodiments, the REINFORCE algorithm can be used to approximate and optimize the gradient. The REINFORCE algorithm is a reinforcement learning algorithm based on policy gradient. It estimates the gradient by sampling trajectories and using the action and reward information in the trajectories to update the policy parameters, so that the policy is gradually optimized in the direction of obtaining higher rewards. It is a model-free reinforcement learning algorithm that does not require prior knowledge of the state transition probability of the environment, but directly learns the strategy from the interaction with the environment. Optionally, the efficiency and stability of REINFORCE algorithm training can be further improved by introducing baselines, importance sampling, entropy regularization, etc.

[0049] The above is the first round of adversarial training. By repeating the second and third stages and performing multiple rounds of attack and defense, a detector with strong generalization ability can be trained.

[0050] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment and the computing device embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0051] Figure 6 FIG. 1 is a schematic block diagram of a machine learning device for image processing according to some embodiments of the present specification. Figure 6 As shown, the apparatus 600 includes: a first detector updating unit 602 , a forger updating unit 604 , and a second detector updating unit 606 .

[0052] In some embodiments, the first detector update unit 602 is configured to update the detector based on the real face sample set and a first fake face sample set generated by multiple fake parameters in the configuration pool of the forger; the forger update unit 604 is configured to update the forgery parameters in the configuration pool of the forger based on the updated detector; the second detector update unit 606 is configured to update the detector again based on the real face sample set and a second fake face sample set generated by the fake parameters in the configuration pool of the updated forger.

[0053] It should be noted that the reference Figures 1 to 5 Further actions or steps shown can be performed by Figure 6 For example, the device 600 may include more modules or units to implement the actions or steps described above, or Figure 6 Some of the units or modules shown may be further configured to implement the actions or steps described above, which will not be repeated here.

[0054] Figure 7 Schematic block diagram of an example device 700 that can be used to implement some embodiments of the present specification is shown. Figure 7As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 702 or computer program instructions loaded from a storage unit 706 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0055] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0056] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as method 200. For example, in some embodiments, the method 200 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the method 200 described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the method 200 in any other appropriate manner (e.g., by means of firmware).

[0057] In some embodiments, the methods and processes described above may be implemented as a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present specification.

[0058] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0059] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0060] The computer program instructions for performing the operation of each embodiment of this specification can be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data or source code or object code written in any combination of one or more programming languages, and the programming languages ​​include object-oriented programming languages, and conventional procedural programming languages. Computer-readable program instructions can be executed completely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on the remote computer, or executed completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer by any type of network-including local area network (LAN) or wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect by the Internet). In certain embodiments, by using the state information of computer-readable program instructions to customize electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGAs) or programmable logic arrays (PLAs), the electronic circuits can execute computer-readable program instructions, thereby realizing the various aspects of this specification.

[0061] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0062] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0063] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The embodiments of the present specification have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

[0065] The above are only optional embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A machine learning method for image processing, the method being used in a training system including a detector and a forger, the method comprising performing the following actions at least once: updating the detector based on a real face sample set and a first forged face sample set generated by a plurality of forged parameters in a configuration pool of the forger; Based on the updated detector, updating the forgery parameters in the configuration pool of the forger; as well as The detector is updated again based on the real face sample set and a second forged face sample set generated by the updated forged parameters in the configuration pool of the forger.

2. The method according to claim 1, wherein generating the first forged face sample set comprises: Initializing fake parameters in the configuration pool; as well as The first forged face sample set is generated based on the real face sample set and the initialized forgery parameters.

3. The method according to claim 1, wherein the type of the forged parameters in the configuration pool includes at least one of the following: Forged objects; Fake area; Type of fusion; and Fusion ratio.

4. The method of claim 1, wherein the detector is trained to perform the following operations: Determine a first loss based on a binary classification result of the input face sample; and A second loss is determined based on the reconstructed forgery parameters of the forged face sample in the binary classification result.

5. The method of claim 4, wherein updating the detector comprises: Under the condition that the forged parameters in the configuration pool are fixed, the network parameters of the detector are updated based on the first loss and the second loss.

6. The method of claim 4, wherein updating the counterfeiter comprises: When the parameters of the detector are fixed, the forgery parameters of the configuration pool of the forger are updated based on the first loss and the second loss.

7. The method of claim 1, wherein the detector comprises a deep fake detection model.

8. A machine learning device for image processing, comprising: A first detector updating unit, configured to update the detector based on a real face sample set and a first forged face sample set generated by a plurality of forged parameters in a configuration pool of the forger; a counterfeiter updating unit configured to update counterfeiting parameters in the configuration pool of the counterfeiter based on the updated detector; as well as The second detector updating unit is configured to update the detector again based on the real face sample set and a second forged face sample set generated by the forged parameters in the configuration pool of the updated forger.

9. A computing device comprising: at least one processing unit; At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the computing device to perform the method as claimed in any one of claims 1 to 7.

10. A computer program product comprising machine executable instructions which, when executed by a device, cause the device to perform the method of any one of claims 1 to 7.

11. A training system comprising: A forger, the forger comprising a configuration pool, the configuration pool comprising a plurality of forgery parameters, the forger being used to generate a forged face sample set based on a real face sample set and the forgery parameters in the configuration pool; as well as The detector is used to perform a binary classification operation on an input face sample and to perform a forged parameter reconstruction operation on a forged face sample in the input face sample.

Images