Image authenticity verification method and device based on a dark watermark and a physical distortion feature

Abstract

The application discloses a kind of based on dark watermark and physical distortion feature image authenticity verification method and device.The method comprises: generating embedding first dark watermark electronic template file, first dark watermark can be detected in electronic template, after being subjected to pre-set physical process, its signal attenuation or loss;Receive user uploaded image to be detected, image to be detected includes second dark watermark and all remaining contents of electronic template file except first dark watermark, second dark watermark and the type and text content of first dark watermark are same;Second dark watermark signal intensity ratio is obtained based on the signal intensity of first dark watermark and second dark watermark, and / or, based on the identifiable number of first dark watermark and second dark watermark, the identifiable rate of second dark watermark is obtained.Using the embodiment of the application can improve the accuracy of image authenticity inspection.

Description

Technical Field

[0001] This application relates to the field of image analysis technology, and also to image watermarking and image anti-counterfeiting detection technology, and in particular to an image authenticity verification method, device, equipment, program product and storage medium based on dark watermark and physical distortion features. Background Technology

[0002] With the acceleration of digitalization, the authenticity of paper documents bearing official seals uploaded by users is frequently verified in numerous business scenarios such as human resources, academic qualification verification, and financial account opening. Traditional manual review methods are inefficient and susceptible to subjective factors, while simple image-text consistency checks are insufficient to combat increasingly sophisticated digital forgery techniques. In particular, when attackers obtain official document templates, they can easily synthesize visually indistinguishable stamped documents using common image editing software. Such forgery poses a serious challenge to existing automated verification systems.

[0003] Currently, several technical approaches exist in the industry for protecting and verifying document authenticity, but all have inherent limitations. While digital watermarking technology can embed hidden information for electronic document traceability, its design often focuses on ensuring stable extraction even after printing and scanning. Physical anti-counterfeiting technologies rely on special paper, ink, or printing processes, resulting in high costs and complex procedures, making them difficult to popularize in ordinary printing scenarios for individuals or small and medium-sized organizations. Analysis methods based on general image features can detect some traces of tampering, but their ability to identify targeted attacks such as partially forged seals or text on genuine templates is limited. Electronic seal systems can ensure the authenticity of the stamping process, but they require relevant organizations to pre-connect to a dedicated system, altering users' ingrained "print, stamp, photograph" operating habits and limiting their deployment applicability.

[0004] Therefore, there is an urgent need for a non-intrusive, automated, and low-cost solution that can accurately distinguish between the formal process of "digitizing documents after they are actually printed and physically stamped" and the tampering behavior of "digitally forging documents directly on electronic templates". Summary of the Invention

[0005] In view of this, embodiments of this application provide an image authenticity verification method, apparatus, electronic device, computer-readable storage medium, and computer program product based on dark watermark and physical distortion features, solving at least one technical problem.

[0006] This application provides an image authenticity verification method based on hidden watermarks and physical distortion features, comprising: generating an electronic template file embedding a first hidden watermark, wherein the first hidden watermark is detectable in the electronic template and its signal is attenuated or lost after a preset physical process; receiving a user-uploaded image to be detected, wherein the image to be detected includes a second hidden watermark and all other content of the electronic template file except for the first hidden watermark, wherein the second hidden watermark has the same type and text content as the first hidden watermark; obtaining a signal strength ratio of the second hidden watermark based on the signal strength of the first hidden watermark and the second hidden watermark, and / or obtaining a recognition rate of the second hidden watermark based on the number of recognizable elements of the first hidden watermark and the second hidden watermark; extracting at least one physical distortion feature from the image to be detected, and calculating a physical distortion score based on the corresponding physical distortion feature, and calculating a total physical distortion score based on each physical distortion score; determining whether the image to be detected is obtained through a genuine physical process based on the watermark signal strength ratio and / or recognition rate, and the total score of the physical distortion features, and if so, determining that the image to be detected is a genuine image.

[0007] Optionally, according to the method of the embodiments of this application, the first dark watermark and the second dark watermark are at least one of the following: text dark watermark, which embeds text content in a low grayscale, slanted, and repetitive tiling manner; frequency domain dark watermark, which is a modulated digital signal embedded in an electronic template file or a preset frequency domain position.

[0008] Optionally, according to the method of this application embodiment, if the first dark watermark and the second dark watermark are text dark watermarks, the identifiability rate of the second dark watermark is obtained based on the identifiable number of the first dark watermark and the second dark watermark, including: using OCR to identify the number of the first dark watermark and the second dark watermark, and counting the number of identifiable second dark watermarks and the first dark watermark; calculating the ratio of the number of second dark watermarks to the number of first dark watermarks to obtain the identifiability rate.

[0009] Optionally, according to the method of this application embodiment, if the first dark watermark and the second dark watermark are frequency domain dark watermarks, the signal intensity ratio of the second dark watermark is obtained based on the signal intensity of the first dark watermark and the second dark watermark, including: calculating the extracted signal intensity ratio of the first dark watermark and the second dark watermark to obtain the watermark signal intensity ratio.

[0010] Optionally, according to the method of this application embodiment, the physical distortion features include at least one of printing halftone features, lens frequency domain notch features, and color channel correlation features.

[0011] Optionally, according to the method of the embodiments of this application, calculating physical distortion scores based on the physical distortion features and calculating a total physical distortion score based on each physical distortion score includes: determining whether the physical distortion features of the image to be detected meet preset conditions, and converting the determination result into a binary determination value to obtain the physical distortion score of the halftone features; and calculating the total physical distortion score by weighted summation based on the weights of each physical distortion feature and the physical distortion score.

[0012] Optionally, according to the method of this application embodiment, determining whether the image to be detected is obtained from a real physical process based on the watermark signal intensity ratio and / or identifiability rate, and the total score of the physical distortion features, includes: calculating a dark watermark feature score based on the watermark signal intensity ratio and / or identifiability rate; calculating a comprehensive score based on the watermark feature score, the total score of the physical distortion features, and their weights; comparing the comprehensive score with a preset threshold, and outputting the authenticity determination result of the image to be detected based on the comparison result.

[0013] This application provides an image authenticity verification device based on dark watermarks and physical distortion features, comprising: an embedding module for generating an electronic template file embedding a first dark watermark, wherein the first dark watermark is detectable in the electronic template and its signal is attenuated or lost after a preset physical process; a receiving module for receiving an image to be detected uploaded by a user, wherein the image to be detected includes a second dark watermark and all other content of the electronic template file except for the first dark watermark, wherein the second dark watermark has the same type and text content as the first dark watermark; and a calculation module for obtaining the signal strength ratio of the second dark watermark based on the signal strength of the first dark watermark and the second dark watermark, and / or obtaining the identifiability rate of the second dark watermark based on the identifiable number of the first dark watermark and the second dark watermark. An extraction module is used to extract at least one physical distortion feature from the image to be detected, calculate a physical distortion score based on the physical distortion feature, and calculate a total physical distortion score based on each physical distortion score; a judgment module is used to determine whether the image to be detected is obtained from a real physical process based on the watermark signal intensity ratio and / or identifiability rate, and the total score of the physical distortion features; if so, the image to be detected is a real image.

[0014] This application provides an electronic device, which includes a processor and a memory storing computer program instructions; the electronic device executes the computer program instructions to implement the method described above.

[0015] This application provides a computer program product, which includes computer program instructions that, when executed, implement the method described above.

[0016] This application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the method described above.

[0017] This application's embodiments achieve efficient and accurate verification of image authenticity by fusing dark watermark signal attenuation with multi-dimensional physical distortion features. Without altering users' existing operating habits or hardware environment, it enables automated, low-cost, and non-invasive identification of images obtained from real physical processes. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly described below.

[0019] Figure 1 This is a schematic diagram of the system architecture of an embodiment of this application.

[0020] Figure 2 This is a flowchart of an image authenticity verification method based on dark watermarking and physical distortion features according to an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the image authenticity verification process based on dark watermark and physical distortion features according to an embodiment of this application.

[0022] Figure 4 This is a flowchart of the printing halftone feature detection process according to an embodiment of this application.

[0023] Figure 5 This is a flowchart of lens frequency domain notch feature detection according to an embodiment of this application.

[0024] Figure 6 This is a flowchart of the color channel correlation feature detection process according to an embodiment of this application.

[0025] Figure 7 This is a schematic structural block diagram of an image authenticity verification device based on dark watermark and physical distortion features according to an embodiment of this application.

[0026] Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0027] The principles and spirit of this application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided to make the principles and spirit of this application clearer and more thorough. The exemplary embodiments provided herein are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0028] Embodiments of this application relate to terminal devices and / or servers. Implementations of this application can be a system, terminal, device, method, computer-readable storage medium, or computer program product, and can be specifically implemented as entirely hardware, entirely software, or a combination of hardware and software. Figure 1 This diagram illustrates a system architecture according to an embodiment of the present application, including a terminal device 102 and a server 104. The terminal device 102 may include at least one of the following: a smartphone, tablet computer, laptop computer, desktop computer, smart TV, various wearable devices, augmented reality (AR) devices, virtual reality (VR) devices, etc. A client application, such as an app, mini-program, or browser-based client, can be installed on the terminal device 102. Users can input commands through the client, and the terminal device 102 can send request information containing the commands to the server 104. Upon receiving the request information, the server 104 performs corresponding processing and returns the processing result information to the terminal device 102. The server 104 may be a local server or a cloud server, and may be a single server or a server cluster, etc.

[0029] In this document, the terms "first," "second," "third," etc., are used only to distinguish one entity (or operation) from another in textual description, and do not require or imply any sequential order between these entities (or operations).

[0030] Figure 2 The diagram illustrates a flowchart of an image authenticity verification method based on dark watermarking and physical distortion features according to an embodiment of this application. The method includes the following steps: S101: Generate an electronic template file embedding the first dark watermark. The first dark watermark can be detected in the electronic template, and its signal is attenuated or lost after passing through a preset physical process. S102: Receive the image to be detected uploaded by the user. The image to be detected includes the second dark watermark and all other contents of the electronic template file except for the first dark watermark. The second dark watermark is the same as the first dark watermark in terms of type and text content. S103: Based on the signal strength of the first dark watermark and the second dark watermark, obtain the signal strength ratio of the second dark watermark, and / or, based on the number of identifiable items of the first dark watermark and the second dark watermark, obtain the identifiability rate of the second dark watermark. S104: Extract at least one physical distortion feature from the image to be detected, calculate the physical distortion score based on the corresponding physical distortion feature, and calculate the total physical distortion score based on each physical distortion score; S105: Based on the watermark signal strength ratio and / or identifiability, and the total score of physical distortion features, determine whether the image to be detected is obtained from a real physical process. If so, the image to be detected is a real image.

[0031] This application provides an image authenticity verification method based on dual evidence chain cross-verification. Its core processing revolves around a reverse logic: traditional digital watermark anti-counterfeiting relies on the stable existence of the watermark as evidence of authenticity, while this application utilizes a fragile dark watermark that is extremely sensitive to physical processes. Its significant attenuation or loss serves as key evidence that the image has undergone a genuine "printing, stamping, and photographing" physical process. Simultaneously, the solution systematically captures the multi-dimensional, coherent distortion traces inevitably introduced by this physical process as auxiliary verification, forming a difficult-to-forge dual anti-counterfeiting barrier.

[0032] In this embodiment, an electronic template file pre-embedded with a first watermark is required. This electronic template file also includes pre-set text or images; for example, an electronic template file for an employment certificate includes fixed template text information, requiring the user to supplement personal identification information and affix an official seal. In actual practice, after receiving the electronic template, the user needs to supplement the required information in the electronic template file, print out the electronic document, and affix their official seal.

[0033] Furthermore, in this application, a dark watermark refers to specific information embedded in an image in a way that is difficult to detect with the naked eye. For example, it might involve repeatedly tiling very light gray text onto a background, or modulating a specific signal into the image's frequency domain (i.e., the frequency distribution domain obtained through mathematical transformation of the image). Its key characteristic, vulnerability, means that while the watermark can be reliably detected by specialized algorithms in the digital domain, it is severely damaged once subjected to actual physical printing (ink diffusion, resolution loss) and subsequent photography (uneven lighting, lens blur, sensor noise). After obtaining the template, the user prints, stamps, and photographs it according to the standard procedure, uploading it as one of the images to be detected. In this application, the images to be detected include real images that have undergone the actual process and forged images obtained using techniques such as Photoshop without undergoing the actual process. Both types of images contain dark watermarks, and these dark watermarks in these two types of images are referred to as the second dark watermark.

[0034] To determine whether an image under test is a genuine image that has undergone a real process or a forged image, the dark watermark in the image needs to be verified and analyzed. Specifically, the signal strength ratio and / or recognizability of the second dark watermark relative to the original first dark watermark are used as one of the verification criteria to determine whether the image under test is genuine.

[0035] In addition to watermark detection, this application also introduces physical distortion feature detection to improve detection accuracy. The actual printing and photographing process leaves a series of intrinsically related physical information in the image. This application extracts and analyzes physical distortion features, such as printing halftone (periodic dot patterns produced by a printer), frequency domain notch (signal attenuation at specific frequencies by a scanner or camera imaging system), and color channel correlation distortion (color difference caused by natural lighting), to determine whether the image under test has undergone the actual process. Each feature is quantified and scored, and the scores are finally aggregated into a total physical distortion feature score.

[0036] Ultimately, by comprehensively evaluating the degree of watermark attenuation (low ratio indicates true) and the degree of physical distortion features (high degree indicates true), the system determines whether the image being tested is obtained from a genuine physical process. By employing both dark watermarking and physical distortion features for image anti-counterfeiting detection, even if a forger attempts to imitate a genuine process by removing the watermark through post-processing, they will be detected because the naturally occurring physical distortion features will be simultaneously erased or distorted. Conversely, if physical distortion is carefully fabricated, it becomes difficult to explain why a fragile watermark does not experience the expected severe attenuation, thus leading to detection. This dual constraint significantly increases the complexity of forgery and the accuracy of detection.

[0037] This application embodiment constructs a dual anti-counterfeiting logic system by cross-verifying the evidence of the reverse attenuation of fragile watermarks with the evidence of the positive appearance of multi-dimensional physical distortion traces. This not only significantly increases the technical threshold and cost of counterfeiting and effectively identifies direct electronic tampering, but also systematically identifies counterfeit images that have undergone complex post-processing and attempt to simulate physical processes. Thus, without changing the user's existing operating habits and hardware environment, it achieves high-precision and automated verification of the authenticity of the digitization process of paper documents.

[0038] In some embodiments of this application, optionally, the first dark watermark and the second dark watermark are at least one of the following dark watermarks: Text watermarking is embedded in the text content using low grayscale, slanted, and repeated tiling. Frequency domain dark watermarking is a modulated digital signal embedded in an electronic template file or at a preset frequency domain location.

[0039] In this application embodiment, the first and second dark watermarks are of the same type. In some embodiments, optionally, the first and second dark watermarks are text dark watermarks. These text dark watermarks are embedded in the electronic template file with extremely low grayscale values ​​(e.g., a light gray close to pure white), a certain tilt angle, and a repeating pattern of microtext across the entire background area. The low grayscale value is located at the edge of the color gradation reproduction capability of most ordinary office printers. When the printer converts the digital grayscale value into ink dot distribution, it is very easy to lose or severely distort this extremely weak brightness difference, causing the information on the printed material to almost disappear. The tilt angle breaks the directional consistency with the document text or common image edges, increasing the difficulty of completely separating it through simple image filtering operations. The repeating pattern ensures that even if local reflections, obstructions, or deformations occur during shooting, the system can still collect enough watermark samples from other areas for statistical identification. When a user takes a picture of a paper document with such an embedded watermark, the automatic exposure, dynamic range compression, and noise of the mobile phone camera will further contaminate the already weak signal. Therefore, during the verification phase, when calculating the recognizable rate using optical character recognition technology, images from the real process will return a very low value because the watermark has been almost completely erased in the physical world; while digitally forged images still have their watermark data in a complete and clear digital form, so their recognizable rate is significantly higher.

[0040] In some embodiments of this application, optionally, the first and second dark watermarks are frequency domain dark watermarks. Any digital image can be converted to the frequency domain for analysis through mathematical transformations (such as discrete Fourier transform). Different positions in the frequency domain image correspond to different frequencies of texture and contour information in the original image. Among them, the mid-to-high frequency part usually corresponds to the details and edges of the image. This application embeds a modulated digital signal (e.g., a specific pseudo-random sequence) representing the watermark information into these preset mid-to-high frequency domain positions. The purpose of modulation is to make the watermark signal better integrate into the frequency characteristics of the host image and improve its concealment. The mid-to-high frequency region is chosen for embedding because the actual physical printing (ink diffusion causes edge blurring) and subsequent digital shooting (lens optical diffraction, sensor low-pass filtering) are essentially low-pass filtering processes, which will inevitably and significantly attenuate the high-frequency information of the image. Therefore, after going through this physical chain, the watermark signal embedded in these frequency bands will be synchronously and severely weakened. At the detection end, the algorithm can quantify this attenuation by calculating the ratio (signal strength ratio) of the watermark signal intensity extracted from the image to be detected to the original embedded intensity. The real process results in a very low ratio, while digitally forged images, which do not undergo this physical low-pass filtering, retain a robust frequency domain watermark signal and have a higher ratio.

[0041] In some embodiments of this application, optionally, both the first and second watermarks can include both text watermarks and frequency domain watermarks, that is, text watermarks and frequency domain watermarks can be embedded in the electronic template file at the same time.

[0042] The textual or frequency-domain watermarking in this application provides solid, specific, and implementable technical support for the evidence chain of watermark signal attenuation, jointly ensuring the effectiveness and reliability of the first layer of verification.

[0043] Optionally, in some embodiments of this application, if the first and second dark watermarks are text watermarks, the identifiability rate of the second dark watermark is obtained based on the identifiable number of the first and second dark watermarks, including: The number of first and second watermarks was identified using OCR, and the number of identifiable second and first watermarks was counted. The recognition rate is obtained by calculating the ratio of the number of second dark watermarks to the number of first dark watermarks.

[0044] In this embodiment, OCR recognition refers to optical character recognition. The system first uses OCR technology to identify and count the first watermark embedded in the original electronic template file, thereby determining the total number of pre-defined watermark characters that can be fully recognized under ideal digital conditions. Subsequently, upon receiving the image uploaded by the user, the system runs the OCR recognition process again in the corresponding area, attempting to extract and count the number of remaining second watermark characters. Due to the actual printing and photographing process, these characters, embedded in extremely low grayscale and at an angle, have strokes that become discontinuous due to the diffusion and merging of ink dots from the printer. The contrast is further reduced due to uneven lighting and automatic gain control in the shooting environment, resulting in a large number of characters becoming unreliable for the OCR engine. Finally, a quantified recognition rate is obtained by calculating the ratio of the number of recognizable second watermark characters to the total number of recognizable first watermark characters.

[0045] Optionally, in some embodiments of this application, if the first and second dark watermarks are frequency domain dark watermarks, the signal intensity ratio of the second dark watermark is obtained based on the signal intensity of the first and second dark watermarks, including: The signal intensity ratio of the extracted first dark watermark and the second dark watermark is calculated to obtain the watermark signal intensity ratio.

[0046] In this embodiment, if the dark watermark is a frequency domain dark watermark, a frequency domain transformation of the image is required using Discrete Fourier Transform. In the frequency domain, the image is decomposed into a combination of signals of different frequencies: low-frequency components correspond to large smooth areas and overall brightness in the image, while high-frequency components correspond to fine details, sharp edges, and textures in the image. This application embeds the watermark information (a specific digital signal) into preset mid-to-high frequency positions using modulation techniques.

[0047] During the verification phase, the system performs a reverse operation. For the original electronic template used as a reference, the algorithm extracts the embedded first dark watermark signal from a preset position in its frequency domain and calculates its energy or amplitude as the original signal strength. For the user-uploaded image to be detected, after preprocessing such as image alignment, the residual second dark watermark signal is extracted from the exact same frequency domain coordinate position. If the image is the product of a real physical process, its high-frequency components have been severely attenuated, and the extracted watermark signal will be very weak; conversely, if it is a digitally forged image, its high-frequency details have been artificially preserved or have not been affected by real physical filtering, and the extracted watermark signal will still be relatively strong. Finally, by calculating the ratio of the second dark watermark signal strength to the first dark watermark signal strength, a normalized signal strength ratio is obtained. A ratio close to 0 indicates that the signal is almost completely attenuated, strongly pointing to a real physical process; a ratio close to 1 indicates that the signal is well preserved, highly suggesting that it is a digital forgery.

[0048] In some embodiments of this application, the physical distortion features optionally include at least one of the following: printing halftone features, lens frequency domain notch features, and color channel correlation features.

[0049] Halftone characteristics in printing are directly related to the printing process. When a digital image is printed onto paper using a regular inkjet or laser printer, the device uses a series of tiny, regularly arranged ink dots (halftones) to simulate continuous color and grayscale changes; this process is called halftone. This microscopically formed periodic dot pattern is a unique imprint left on the image by the printer's physical mechanism. During verification, by performing a frequency domain transformation (such as a Fourier transform) on the image, significant peaks corresponding to this periodicity can be detected in the frequency spectrum. Genuine printed documents exhibit this regular frequency characteristic, while counterfeit images synthesized directly from electronic templates, whose pixel structure is digitally synthesized, typically lack this regular periodic pattern generated by the physical printing mechanism.

[0050] The notch characteristic in the frequency domain is related to the digitization process of shooting or scanning. Every optical imaging device (such as a mobile phone camera or flatbed scanner) has an imperfect lens assembly, sensor, and internal signal processing circuitry; each acts like a unique filter, processing the input light signal. This processing characteristic manifests in the frequency domain as an abnormally weak signal response at certain specific frequencies, forming what is known as a notch. In a real process, when paper documents are photographed, they inevitably undergo filtering by the physical system of the imaging device, leaving the notch characteristic of that device in the image spectrum. Forged images, if directly generated or altered by software, lack this specific physical processing step; their spectrum is either too clean or exhibits an unnatural frequency response pattern inconsistent with common imaging devices. Analyzing this characteristic can serve as evidence to determine whether an image has been digitized using a specific physical imaging device.

[0051] Color channel correlation is related to the interaction between ambient lighting and the shooting process. When natural light shines on a paper document and is captured by a camera sensor, this process causes systematic and interrelated changes in the brightness values ​​of the RGB color channels of the image. For example, under lighting of a specific color temperature, the response ratios of the red, green, and blue channels exhibit a natural correlation. Especially after conversion to the Lab color space, which is more in line with human visual perception of brightness changes, the correspondence between its brightness channels and the original template will show smooth and continuous changes due to uneven lighting, shadows, etc. Forged images, especially those with locally altered areas, may have their color information directly copied from other digital sources or manually adjusted, often disrupting this natural correlation and gradation pattern between global color channels caused by the real physical lighting environment. This may manifest as abnormal color casts or abrupt changes in the statistical characteristics of color in local areas. Examining this characteristic can help determine whether the image's color information originated from a genuine physical shooting process under natural lighting conditions.

[0052] The physical distortion features provided in this application's embodiments corroborate the same physical event from different perspectives (printing mechanism, device fingerprint, light interaction). This multi-dimensional and systematic feature analysis improves the accuracy of detection.

[0053] In some embodiments of this application, optionally, calculating a physical distortion score based on the corresponding physical distortion features, and calculating a total physical distortion score based on each physical distortion score, includes: Determine whether the physical distortion features of the image to be detected meet the preset conditions, and convert the determination result into a binary determination value to obtain the physical distortion score of the halftone features; The total physical distortion score is calculated by weighted summation based on the weights of each physical distortion feature and the physical distortion score.

[0054] In this embodiment, for each detected physical distortion feature, such as a printed halftone feature, the system makes a judgment based on a preset threshold condition derived from physical laws and experimental data. For example, when detecting halftone, the intensity of the periodic pattern is calculated through Fourier spectrum analysis. If its variance is significantly greater than the mean (e.g., more than twice), it is determined to meet the true printing feature, and this judgment result is converted into a binary judgment value, usually 1 to indicate that the feature exists (supports true), and 0 to indicate that it does not exist (does not support true or is doubtful).

[0055] After obtaining the binary judgment values ​​(i.e., the physical distortion scores of each feature) for each feature, the system assigns appropriate weights to each feature based on its reliability and importance in identifying the real physical process, and calculates the final physical distortion score through weighted summation. For example, printing halftone features can be given a higher weight (e.g., 0.3) because the printing characteristics produced by printing are difficult to perfectly simulate; while color channel correlation can be given a slightly lower weight (e.g., 0.2) because lighting effects can be approximated under certain conditions. The result of the weighted summation is a value between 0 and 1, which intuitively represents the degree to which the image under test conforms to the real physical distortion pattern: the higher the total score, the more high-weight physical features are judged to exist, and the greater the possibility that the image comes from a real process; conversely, the lower the total score, the less coherent the physical traces in the image are, and the higher the possibility of forgery.

[0056] In some embodiments of this application, optionally, determining whether the image to be detected is obtained from a real physical process is based on the watermark signal intensity ratio and / or identifiability rate, and the total score of physical distortion features, including: Calculate the dark watermark feature score based on the watermark signal strength ratio and / or recognizability. A comprehensive score is calculated based on watermark feature scoring, physical distortion feature total score, and the weights of the two. The comprehensive score is compared with a preset threshold, and the authenticity judgment result of the image to be detected is output based on the comparison result.

[0057] In this embodiment, if the first dark watermark is either a text dark watermark or a frequency domain dark watermark, the dark watermark feature score is the corresponding identifiability rate or signal strength ratio. If the first dark watermark is both a text dark watermark and a frequency domain dark watermark, the dark watermark feature score is the weighted sum of the signal strength ratio and the identifiability rate, where a higher score indicates less watermark attenuation, i.e., a greater likelihood of forgery. The total score of the physical distortion feature represents the richness of traces in the image that conform to real physical laws; a higher total score indicates stronger authenticity.

[0058] The overall score is calculated by weighting and summing the scores for both the dark watermark feature and the physical distortion feature. The weighting (e.g., watermark feature weight α = 0.4, physical distortion feature weight β = 0.6) reflects the perceived reliability of the two types of evidence. For example, in situations with complex lighting conditions that may affect physical characteristics, the weight of dark watermark evidence might be appropriately increased; conversely, in scenarios using high-precision printers or where watermark attenuation is highly predictable, physical distortion evidence can be given a higher weight. This configurable weighting mechanism allows the system to flexibly adapt to different business environments and security level requirements.

[0059] Finally, by comparing the calculated comprehensive score with one or more preset thresholds determined through training on a large number of real and fake samples, the system outputs the final judgment result. For example, a low threshold (e.g., 0.3) and a high threshold (e.g., 0.7) can be set. If the comprehensive score is below 0.3, it means that the watermark has severely decayed and physical traces are obvious, and the system can determine it as real with high confidence; if it is above 0.7, it means that the watermark is well preserved and lacks coherent physical traces, and can determine it as fake with high confidence; scores in the middle range will result in a conclusion suggesting manual review. This threshold comparison mechanism, especially the introduction of gray-scale decision intervals, greatly improves the practicality and reliability of the system, avoids making arbitrary binary judgments when evidence is ambiguous, and controls the final risk within an acceptable range.

[0060] The implementation methods and advantages of the embodiments of this application have been described above through multiple examples. The specific processing procedures of the embodiments of this application are described in detail below with reference to specific examples.

[0061] Figure 3 This is a schematic diagram illustrating the processing steps of an image authenticity verification method based on dark watermarking and physical distortion features, according to an embodiment of this application. Combined with... Figure 3 As shown, the image authenticity verification method based on dark watermarking and physical distortion features of this application includes the following steps: Step 1: Generate a template file with standard style content.

[0062] Step 2: Embed a hidden watermark in the template file. The type of hidden watermark can be any of the following two types, or both: (1) Text hidden watermark: Select a text watermark area in the template (it is recommended to use the entire background of the document as the text watermark area) and embed a piece of micro text (such as XX platform) in a very low grayscale (such as RGB 252,252,252), tilt angle, and repeating tiling method. The watermark text is barely visible on the screen, but it will be ignored or severely lost when printed by a regular printer. (2) Frequency domain hidden watermark: Embed a modulated fixed digital signal (which can be a random string or fixed text content) in the preset high frequency region of the discrete Fourier transform (DFT) domain of the template image.

[0063] Step 3: Receive the image to be tested uploaded by the user.

[0064] Step 4: The platform preprocesses the document images uploaded by users, such as cropping the images appropriately, removing non-document backgrounds, and aligning them with the original template document.

[0065] Step 5: Dual feature verification.

[0066] (1) Dark watermark feature detection: The text dark watermark detection process includes: locating the text watermark area in the preprocessed image and extracting the watermark text content through OCR recognition technology. The number of successfully identified embedded watermark texts is counted, and the watermark text recognition rate is calculated as: number of recognizable watermark texts ÷ total number of original embedded watermark texts. The recognition rate ranges from [0,1]. For genuine printed images, the watermark recognition rate is usually low (empirical threshold such as <0.3); forged images maintain a high recognition rate. The frequency domain dark watermark detection process includes: performing a discrete Fourier transform on the preprocessed image and attempting to extract and decode the watermark signal at a preset frequency domain position. The watermark signal strength ratio is calculated as: extracted signal strength ÷ original embedded signal strength (or calculating the signal-to-noise ratio and normalizing it). The strength ratio is normalized to the range of [0,1]. In genuine printed images, the watermark signal is significantly attenuated, resulting in a low strength ratio; forged images maintain a high strength ratio.

[0067] (2) Physical Distortion Feature Detection: Detects the following three key features of the image. Other features can be added or removed based on the actual application, such as ink diffusion edge features, noise-related features, etc. Feature 1: Printing Feature Detection (Halftone Detection), analyzes whether the image has periodic dot patterns in a specific frequency band, used to detect physical features generated by the printer during the printing process. Combined with... Figure 4The process includes the following steps: receiving the uploaded image to be detected and performing document cropping and grayscale preprocessing. Document cropping includes identifying document region boundaries in the image using edge detection or contour analysis algorithms, cropping to remove the background, and retaining only the clean document content area. Grayscale conversion and contrast enhancement include converting the color image to a single-channel grayscale image, preserving brightness information, and eliminating color interference for detection. To enhance the detection effect of periodic features, adaptive contrast adjustment can be performed. Then, the spatial domain image is converted to a frequency domain representation using Fourier transform, shifting the zero-frequency component to the center of the spectrum for easier analysis, calculating the amplitude spectrum, and normalizing the numerical range using logarithmic transformation. Next, a central region is extracted for analysis: a square region of fixed size is extracted with the image center as the origin (generally 5% of the minimum image dimensions, e.g., 100×100). This region corresponds to the low-frequency components of the image and contains the main information of the periodic pattern. Finally, the variance and mean of the pixel values ​​in the extracted region are calculated, where the variance reflects the degree of brightness fluctuation, and the mean represents the average brightness. The judgment is made by using twice the empirical threshold. When the variance is significantly greater than the mean (e.g., more than twice the mean), it indicates the existence of a clear periodic pattern and is judged as having printing characteristics; otherwise, it is judged as not having them.

[0068] Feature 2: Frequency Domain Notch Detection. This feature identifies the device fingerprint characteristics of image processing by analyzing the image's spectral characteristics to detect the presence of specific frequency notch patterns generated by the built-in signal processors of hardware devices such as scanners and printers. This feature serves as important evidence in determining whether an image has been physically printed or re-digitized (scanned or photographed). Combined with... Figure 5 As shown, the detection process includes: inputting the original template image and the image to be detected for comparative analysis of device fingerprint features; preprocessing the image using document cropping alignment and data standardization, specifically including (1) background cropping: identifying the document region boundary in the image to be detected through edge detection or contour analysis algorithms, cropping to remove the shooting background, and retaining only the pure document content area. (2) geometric coordinate alignment: geometrically aligning the cropped document image to be detected with the template document image, and establishing a stable coordinate correspondence between the two images through affine transformation or feature matching. (3) grayscale conversion and size standardization: converting the two images into single-channel grayscale format respectively, and scaling the image to be detected to the same size based on the template image size (or uniformly adjusting to a preset standard size such as 800×800 pixels). (4) pixel value normalization: normalizing the image pixel values ​​to the 0-1 value range to provide standardized input for subsequent frequency domain analysis.

[0069] Then, spectral analysis and system characteristic extraction are performed. Specifically, Fourier transforms are performed on the two images to obtain their frequency domain representations, and the power spectrum energy distribution is calculated. A one-dimensional energy distribution curve is obtained through radial averaging, and the ratio of the two is calculated to obtain the system frequency response (transfer function). This curve reflects the processing characteristics of the imaging device for each frequency component. Finally, notch feature analysis is performed: on the system transfer function curve, the curve is inverted to find significant peaks (corresponding to the valleys of the original curve), and then the depth of the valley is calculated: the response value at the valley is subtracted from 1.0 to obtain the notch depth. An empirical threshold of 0.3 is used for judgment: if the depth exceeds 0.3, it is determined that a frequency domain notch feature exists; otherwise, it is determined that no notch feature exists.

[0070] Feature 3: Color Channel Correlation: This feature reflects the interaction between lighting and the environment, highlighting how lighting systematically affects color in real-world environments. The image is converted to the Lab color space, and the structural similarity between the uploaded image and the original template in the Luminosity (L) channel is calculated. Real-world shooting results in a moderate decrease in correlation due to variations in lighting; forged images with local modifications may exhibit abnormally high correlation or localized abrupt changes. Combined with... Figure 6 As shown, the detection process includes: inputting the original template image and the image to be detected for comparative analysis of color distortion data; preprocessing the image using document cropping, geometric alignment and size unification, specifically including (1) cropping the background: identifying the document area boundary in the image to be detected through edge detection or contour analysis algorithms, cropping to remove the shooting background, and retaining only the pure document content area. (2) geometric coordinate alignment: geometrically aligning the cropped document image to be detected with the template document image, and establishing a stable coordinate correspondence between the two images through affine transformation or feature matching. (3) size standardization: scaling the image to be detected to the same size based on the size of the template image (or uniformly adjusting to a preset standard size such as 800×800 pixels). Then, converting the two images from RGB color space to Lab color space, focusing on separating the brightness channel (L), which is most sensitive to brightness changes during the printing shooting process. Next, calculating two key data indicators of the brightness channel: mean and variance, where the proportion of mean change reflects brightness shift, and the proportion of variance change reveals drastic changes in contrast / noise. If the variance ratio is significantly greater than the empirical threshold (e.g., >3) and the mean change ratio is less than the empirical threshold (e.g., <0.8), it is determined that there are real shooting lighting characteristics; otherwise, it is determined that there are none.

[0071] The detection results based on the above three features are used for weighted decision fusion. First, the quantized output value of each feature is converted into a binary judgment (0 indicates that it conforms to the real physical features, and 1 indicates that it does not) through a threshold function. Then, the values ​​are weighted and summed according to preset weight coefficients to calculate the total score of the physical distortion feature (interval [0,1]). This score reflects the probability of the image being forged; the higher the score, the more likely it is to be a forged image. The weight allocation example is as follows: Feature 1 (printing feature): weight 0.25; Feature 2 (frequency domain notch filtering): weight 0.25; Feature 3 (color correlation): weight 0.2.

[0072] Step 6: Feature fusion and decision. The dark watermark feature score W∈[0,1] (the higher the score, the greater the possibility of forgery) and the physical distortion feature total score P∈[0,1] (the higher the score, the greater the possibility of forgery) are fused at the decision level to generate the final authenticity judgment. Among them, the fusion strategy can be one of the following two: (1) Rule-weighted fusion: the two features are weighted and summed according to the preset weight coefficients to calculate the comprehensive forgery probability. The comprehensive score S∈[0,1], where S = α×W +β×P (α+β=1.0, such as α=β=0.5). Based on the empirical threshold, a triple decision judgment is made (example thresholds 0.2 and 0.8): when S<0.2: judged as real image (high confidence); when S>0.8: judged as forged image (high confidence); when 0.2≤ S≤0.8: manual review is recommended (medium to low confidence). (2) Machine learning integration: Construct a multidimensional feature vector X=[W, feature1, feature2, feature3], collect labeled samples to train the classification model, and directly output the true classification and confidence level.

[0073] Decision output: The system finally outputs a binary decision (real / fake) and the corresponding confidence score (if machine learning fusion is used).

[0074] Step 7: If the image is determined to be genuine, the process ends normally. If the image is determined to be fake, the process proceeds to manual review.

[0075] Step 8: Manual review completed.

[0076] Step 9: Process ends.

[0077] In summary, the proposed solution has the following advantages: (1) Overall, dual verification ensures high robustness: This application innovatively combines the seemingly contradictory features of the weak existence of fragile watermarks and the appearance of physical distortion features with complementary evidence confirming the same physical event (printing and taking photos). For an attacker to forge, they must simultaneously satisfy: perfectly remove the fragile watermark (simulating the physical process that damages the watermark) and perfectly simulate the coherent physical distortion features. This greatly increases the technical threshold and cost of forgery, forming an anti-counterfeiting effect of "1+1>2". Moreover, it is non-invasive to the existing process: The entire solution is based entirely on image algorithms, requiring no special hardware (such as special paper or dedicated scanners) from the user, nor does it require changing the user's existing printing, stamping, and photo-taking operation process.

[0078] (2) In terms of details, the text-based dark watermarking scheme utilizes the color gradation compression characteristics of ordinary printers and the dynamic range limitations of mobile phone cameras to allow the low-grayscale watermark to naturally attenuate and weaken in the actual physical process. OCR technology is used to detect the recognition rate, which is intuitive and computationally efficient. The frequency domain dark watermarking scheme embeds the watermark into the mid-to-high frequency region, which is sensitive to Gaussian blur, geometric deformation, and resolution reduction. The normal printing and photography process can ensure the reliable attenuation of the watermark signal, and its invisibility provides stronger concealment protection.

[0079] (3) The selection of physical distortion features constitutes a complete physical process evidence chain, which corresponds to key links respectively: halftone features in the printing link, frequency domain notch features in the digitization link, and color correlation features in the illumination link. This systematically forms a full-link anti-counterfeiting detection, avoiding the risk that single-point features are easily bypassed.

[0080] Correspondingly, this application also provides an image authenticity verification device based on dark watermarking and physical distortion features, referencing Figure 7 ,include: The embedding module 110 is used to generate an electronic template file embedding the first dark watermark. The first dark watermark can be detected in the electronic template, and its signal is attenuated or lost after a preset physical process. The receiving module 120 is used to receive the image to be detected uploaded by the user. The image to be detected includes a second dark watermark and all the remaining contents of the electronic template file except for the first dark watermark. The second dark watermark is the same as the first dark watermark in terms of type and text content. The calculation module 130 is used to obtain the signal strength ratio of the second dark watermark based on the signal strength of the first dark watermark and the second dark watermark, and / or to obtain the identifiability rate of the second dark watermark based on the identifiability of the first dark watermark and the second dark watermark. The extraction module 140 is used to extract at least one physical distortion feature from the image to be detected, calculate a physical distortion score based on the physical distortion feature, and calculate a total physical distortion score based on each physical distortion score. The judgment module 150 is used to determine whether the image to be detected is obtained from a real physical process based on the watermark signal strength ratio and / or identifiability rate, and the total score of the physical distortion features. If so, the image to be detected is a real image.

[0081] Based on at least one of the above embodiments, the electronic device in the embodiments of this application may be a user terminal device, a server, other computing devices, or a cloud server. Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. The electronic device may include a processor 601 and a memory 602 storing computer program instructions. The processor 601 reads and executes the computer program instructions stored in the memory 602 to implement the process or function of any of the methods in the above embodiments.

[0082] Specifically, processor 601 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Memory 602 may include a mass storage device for data or instructions. For example, memory 602 may be at least one of the following: a hard disk drive (HDD), read-only memory (ROM), random access memory (RAM), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, universal serial bus (USB) drive, or other physical / tangible memory storage device. Alternatively, memory 602 may include removable or non-removable (or fixed) media. Furthermore, memory 602 may be internal or external to the integrated gateway disaster recovery device. Memory 602 may be non-volatile solid-state memory. In other words, typically memory 602 includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with computer-executable instructions, and when the software is executed (e.g., by one or more processors), it can perform the operations described in the methods of the embodiments of this application.

[0083] As an example, Figure 8The illustrated electronic device may also include a communication interface 603 and a bus 610. The processor 601, memory 602, and communication interface 603 are connected via bus 610 and communicate with each other. Bus 610 may include hardware, software, or both, and may couple components of an online data traffic metering device together. The bus may include at least one of the following: Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), HyperTransport (HT) Interconnect, Industry Standard Architecture (ISA) bus, Infinite Bandwidth Interconnect, Low Pin Count (LPC) bus, memory bus, Microchannel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus, or other suitable bus. Bus 610 may include one or more buses. Although specific buses are described or shown in the embodiments of this application, any suitable bus or interconnection method is contemplated in the embodiments of this application.

[0084] In conjunction with the methods in the above embodiments, this application also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the process or function of any of the methods in the above embodiments.

[0085] This application also provides a computer program product that stores computer program instructions, which, when executed by a processor, implement the process or function of any of the methods described above.

[0086] The flowcharts and / or block diagrams of methods, terminals, systems, and computer program products according to embodiments of this application have been exemplarily described above, and related aspects have been described. It should be understood that each block or combination thereof in the flowcharts and / or block diagrams may be implemented by computer program instructions, by dedicated hardware performing a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine such that these instructions, executed via such processor, enable the implementation of the function / action specified in each block or combination thereof in the flowcharts and / or block diagrams. Such a processor may be a general-purpose processor, a dedicated processor, a special-purpose application processor, or a field-programmable logic circuit.

[0087] The functional blocks shown in the structural block diagrams of this application can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc.; when implemented in software, they are programs or code segments used to perform the required tasks. Programs or code segments can be stored in memory or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. Code segments can be downloaded via computer networks such as the Internet or intranets.

[0088] It should be noted that this application is not limited to the specific configurations and processes described above or shown in the figures. The above descriptions are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the described systems, devices, terminals, modules, or units can be referred to the corresponding processes in the method embodiments, and need not be repeated here. It should be understood that the scope of protection of this application is not limited thereto. Any equivalent modifications or substitutions that can be conceived by those skilled in the art within the scope of the technology disclosed in this application should be covered within the scope of protection of this patent application.

Claims

1. An image authenticity verification method based on dark watermarking and physical distortion features, characterized in that, include: An electronic template file is generated that embeds a first dark watermark. The first dark watermark can be detected in the electronic template, and its signal is attenuated or lost after a preset physical process. The system receives an image to be detected uploaded by a user. The image to be detected includes a second dark watermark and all other contents of an electronic template file except for the first dark watermark. The second dark watermark is the same as the first dark watermark in terms of type and text content. The signal strength ratio of the second dark watermark is obtained based on the signal strength of the first dark watermark and the second dark watermark, and / or the identifiability rate of the second dark watermark is obtained based on the identifiable number of the first dark watermark and the second dark watermark. Extract at least one physical distortion feature from the image to be detected, calculate a physical distortion score based on the physical distortion feature, and calculate a total physical distortion score based on each physical distortion score. Based on the watermark signal strength ratio and / or identifiability, and the total score of the physical distortion features, it is determined whether the image to be detected is obtained from a real physical process. If so, the image to be detected is a real image.

2. The method according to claim 1, characterized in that, The first and second watermarks are at least one of the following: Text watermarking is embedded in the text content using low grayscale, slanted, and repeated tiling. Frequency domain dark watermarking is a modulated digital signal embedded in an electronic template file or at a preset frequency domain location.

3. The method according to claim 2, characterized in that, If both the first and second watermarks are text watermarks, then the recognizability rate of the second watermark is obtained based on the number of recognizable elements of the first and second watermarks, including: The number of first and second watermarks was identified using OCR, and the number of identifiable second and first watermarks was counted. The recognition rate is obtained by calculating the ratio of the number of second dark watermarks to the number of first dark watermarks.

4. The method according to claim 2, characterized in that, If the first and second dark watermarks are frequency domain dark watermarks, then the signal strength ratio of the second dark watermark is obtained based on the signal strengths of the first and second dark watermarks, including: The signal intensity ratio of the extracted first dark watermark and the second dark watermark is calculated to obtain the watermark signal intensity ratio.

5. The method according to claim 1, characterized in that, The physical distortion features include at least one of the following: printing halftone features, lens frequency domain notch features, and color channel correlation features.

6. The method according to claim 1, characterized in that, The physical distortion score is calculated based on the physical distortion features, and the total physical distortion score is calculated based on each physical distortion score, including: Determine whether the physical distortion features of the image to be detected meet the preset conditions, and convert the determination result into a binary determination value to obtain the physical distortion score of the halftone features; The total physical distortion score is calculated by weighted summation based on the weights of each physical distortion feature and the physical distortion score.

7. The method according to claim 1, characterized in that, Based on the watermark signal intensity ratio and / or identifiability rate, and the total score of the physical distortion features, determining whether the image to be detected is obtained from a real physical process includes: Based on the watermark signal intensity ratio and / or identifiability, a dark watermark feature score is calculated. A comprehensive score is calculated based on the watermark feature score, the total score of the physical distortion feature, and the weights of the two. The comprehensive score is compared with a preset threshold, and the authenticity determination result of the image to be detected is output based on the comparison result.

8. An image authenticity verification device based on dark watermarking and physical distortion features, characterized in that, include: The embedding module is used to generate an electronic template file that embeds the first dark watermark. The first dark watermark can be detected in the electronic template, and its signal is attenuated or lost after a preset physical process. The receiving module is used to receive the image to be detected uploaded by the user. The image to be detected includes a second dark watermark and all the remaining content of the electronic template file except for the first dark watermark. The second dark watermark is the same as the first dark watermark in terms of type and text content. The calculation module is used to obtain the signal strength ratio of the second dark watermark based on the signal strength of the first dark watermark and the second dark watermark, and / or, based on the number of identifiable items of the first dark watermark and the second dark watermark, obtain the identifiability rate of the second dark watermark. An extraction module is used to extract at least one physical distortion feature from the image to be detected, calculate a physical distortion score based on the physical distortion feature, and calculate a total physical distortion score based on each physical distortion score. The judgment module is used to determine whether the image to be detected is obtained from a real physical process based on the watermark signal strength ratio and / or identifiability rate, and the total score of the physical distortion features. If so, the image to be detected is a real image.

9. An electronic device, characterized in that, The electronic device is a terminal device or a server. The electronic device includes a processor and a memory storing computer program instructions. When the electronic device executes the computer program instructions, it implements the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, It includes computer program instructions that, when executed, implement the method as described in any one of claims 1-7.

11. A computer-readable storage medium, characterized in that, It stores computer program instructions that, when executed, implement the method as described in any one of claims 1-7.

Images