Card hidden information visual reading method and system based on one-key decoding

Abstract

The present application relates to the technical field of computer vision and image processing, and provides a card hidden information visual reading method and system based on one-key decoding, which comprises the following steps: performing multi-angle sequential lighting through an annular light source array, collecting an angle-image sequence, constructing a digital grating transfer function for calculation domain decoding, performing multi-angle differential enhancement and frequency domain filtering, and performing semantic recognition and comparison verification, and optimizing the lighting parameters through closed-loop feedback, wherein the digital decoding success rate reaches 99.5%, the recognition accuracy reaches 98%, and the single decoding time is less than 1s.

Description

Technical Field

[0001] This invention relates to the field of computer vision and image processing technology, and in particular to a method and system for visually recognizing hidden information on cards based on one-click decoding. Background Technology

[0002] In the field of anti-counterfeiting technology for printing, the technique of using grating encoding to hide graphic or image information on the surface of printed materials has been widely applied. The basic principle of this technology is that by performing halftone screening and color plate separation on the original image, the anti-counterfeiting information to be hidden is embedded into the halftone dot structure of the printed material at a specific angle and frequency. When consumers need to verify the anti-counterfeiting information, a physical grating sheet precisely matched to the encoding parameters must be placed on top of the printed material surface at a specific angle. Through the selective masking and transmission of the halftone dots by the grating, the hidden graphic information is revealed. While this anti-counterfeiting identification method based on physical grating decoding improves the anti-counterfeiting level of printed materials to some extent, it still has many shortcomings in practical applications.

[0003] Chinese patent CN116757908A discloses a one-click intelligent encryption method for printed images. By building an image processing interface in a MATLAB environment, it achieves RGB-to-CMYK color space conversion of the original image, halftone screening, and embedding of hidden information. This method does indeed achieve one-click operation for information embedding at the encryption end, intelligently selecting the optimal embedding angle and position for the encrypted information and outputting the parameters required by the decoding tool. However, the above method mainly addresses the automation issues at the information encoding and embedding ends; at the decoding end, it still relies on the use of physical grating sheets. Specifically, when extracting hidden information, users need to manually create a decoded grating image with specific grayscale values, resolutions, and screening angles in image editing software based on the output decoding tool parameters. This decoded grating is then superimposed on the image containing the hidden information, and the hidden content is determined by observing the visual effect of the superimposed grating.

[0004] The aforementioned physical grating-based decoding method faces several prominent problems in practical industrial applications. First, the fabrication and use of physical gratings require precise control of multiple parameters, such as the screening angle, spatial frequency, and stacking position, demanding a high level of expertise from operators. Ordinary consumers find it difficult to independently perform counterfeit verification. Second, the physical grating decoding process is highly dependent on manual operation and human judgment. Decoding results are affected by factors such as operator experience, observation angle, and ambient lighting, lacking objective and quantitative evaluation standards. Third, and most critically, the physical grating decoding method is fundamentally unsuitable for the demands of batch automated testing scenarios. In fields such as card collecting, ticket anti-counterfeiting, and document verification, large-scale card authentication requires highly efficient and consistent automated testing methods. The method of stacking and observing physical gratings one by one is completely unable to meet the throughput requirements of hundreds or even thousands of cards per hour.

[0005] Furthermore, traditional grating decoding methods have significant shortcomings in detection sensitivity. When the embedded contrast of hidden information is low or the printed material is slightly worn, the hidden pattern revealed by the physical grating often lacks contrast and has blurred edges, making it difficult for the human eye to accurately identify, leading to a significant increase in the probability of missed detections and false positives. It is particularly noteworthy that in the field of collectible cards, anti-counterfeiting cards with hidden information inevitably suffer physical damage such as friction, bending, and oil stains during circulation and collection. This damage can cause localized degradation of the encoding grating structure of the hidden information, further reducing the effectiveness of physical grating decoding. In addition, there are processing errors between different batches of physical grating sheets; even small deviations in grating frequency and angle can lead to significant deterioration in decoding performance. This places extremely stringent requirements on the processing precision and consistency of physical gratings, increasing the cost of mass application. Moreover, the physical grating decoding process lacks the ability to quantitatively evaluate the decoding results. Operators can only subjectively judge whether the hidden pattern is clearly visible by human eyes, and cannot provide objective quantitative indicators such as clarity and completeness. This constitutes a significant limitation in application scenarios requiring test reports. In summary, there is an urgent need for a digital and automated hidden information decoding technology that does not rely on physical gratings. This technology should be able to quickly and accurately restore and identify the hidden information embedded in cards without human intervention, providing technical support for batch automated anti-counterfeiting detection. Summary of the Invention

[0006] To address the technical problem that existing technologies rely on physical gratings for decoding hidden card information and cannot adapt to batch automated detection, this invention provides a visual recognition method and system for hidden card information based on one-click decoding.

[0007] The first aspect of the present invention provides a visual recognition method for hidden card information based on one-click decoding, comprising the following steps: a multi-angle sequential illumination step, wherein a ring light source array is used to perform sequential illumination of the surface of the card to be detected at programmable angles, and the light source units at corresponding positions are activated sequentially according to a preset illumination angle sequence to form a sequential illumination process covering a preset angle range; an angle-image sequence acquisition step, wherein a reflected image of the card surface is acquired synchronously at each illumination angle, and the reflected images acquired at all illumination angles are organized into an angle-image sequence in the order of brightness; a digital grating decoding step, wherein a digital grating transfer function matching the encoding parameters of the hidden card information is constructed, and a digital grating filtering operation is applied to the angle-image sequence to simulate the selective transmission effect of a physical grating and extract the initial decoding result of the hidden information; a hidden information enhancement step, wherein multi-angle differential enhancement processing and frequency domain filtering processing are performed on the initial decoding result to suppress background texture and improve the contrast and clarity of the hidden content; and a semantic recognition and comparison verification step, wherein semantic recognition is performed on the enhanced hidden information image and compared with a standard template to determine the authenticity and integrity of the hidden information, and the verification result is fed back to the multi-angle sequential illumination step to adaptively optimize the illumination angle sequence.

[0008] A second aspect of the present invention provides a visual recognition system for hidden card information based on one-click decoding, comprising: a multi-angle sequential illumination module for performing sequential illumination of programmable angles through a ring light source array; an angle-image sequence acquisition module for simultaneously acquiring reflected images at each illumination angle and organizing them into an angle-image sequence; a digital grating decoding module for applying filtering operations to the angle-image sequence through a digital grating transfer function to extract hidden information; a hidden information enhancement module for performing multi-angle differential enhancement and frequency domain filtering to improve the clarity of the hidden content; and a semantic recognition and comparison verification module for performing semantic recognition and comparison verification and feeding the results back to the illumination module.

[0009] The beneficial effects of this invention are as follows: First, by constructing a computational imaging architecture, a digital replacement of physical gratings is achieved, enabling the decoding and recognition of hidden information on cards without the need for physical grating sheets, thus breaking through the dependence of traditional methods on physical gratings and manual operation; Second, through the coordinated cooperation of multi-angle sequential illumination from a ring light source array and digital grating filtering, fully automatic extraction of hidden information is achieved, with a digital decoding success rate of 99.5%, supporting the recognition of hidden information within an angle range of ±15°; Third, through the cascaded processing of multi-angle differential enhancement and frequency domain filtering, the visibility of low-contrast hidden content is effectively improved, with a hidden content recognition accuracy rate of 98%; Fourth, through closed-loop feedback optimization of the illumination angle sequence based on semantic recognition results, adaptive adjustment of detection parameters is achieved, with a single decoding time of less than 1 second, meeting the requirements for batch automated detection.

[0010] Furthermore, the technical solution of this invention has significant technological advancements compared to existing technologies in the following aspects. At the decoding principle level, existing technologies rely entirely on the optical superposition effect of physical gratings for decoding hidden information on cards. The decoding quality is limited by the processing precision and usage conditions of the physical gratings. This invention, by constructing a digital grating transfer function in the computational domain, transforms the physical optical process into a precisely controllable digital signal processing process, fundamentally eliminating the impact of physical grating processing errors on decoding quality. At the information enhancement level, existing technologies can only present hidden patterns after decoding with the naked eye, failing to digitally enhance low-contrast hidden information. This invention, through cascaded processing of multi-angle differential enhancement and frequency domain filtering, can improve the signal-to-noise ratio of hidden information by more than 10dB, enabling clearly discernible decoding results even with slight wear on the card surface. In terms of automation, the decoding process of existing technologies requires operators to manually place the grating, adjust the angle, and interpret the results by eye. This invention realizes full-link automation from lighting control to result output. With the help of a closed-loop feedback optimization mechanism, it can autonomously complete high-quality decoding and detection without human intervention, providing a practical and feasible technical path for industrial-grade batch anti-counterfeiting detection applications. Attached Figure Description

[0011] Figure 1 This is a flowchart of a method for visually recognizing hidden card information based on one-click decoding, provided in an embodiment of the present invention.

[0012] Figure 2 This is an architecture diagram of a card hidden information visual recognition system based on one-click decoding provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the following embodiments are only for illustrative purposes and do not constitute a limitation on the scope of protection of the present invention. Equivalent transformations and improvements made by those skilled in the art without departing from the concept of the present invention are all within the scope of protection of the present invention.

[0014] like Figure 1 As shown in the figure, the visual recognition method for hidden card information based on one-click decoding provided by this invention has the core idea of ​​replacing the traditional optical decoding process of physical gratings with computational imaging principles. It acquires angle-sensitive reflective image sequences through multi-angle sequential illumination, and then uses a digital grating transfer function to simulate the filtering effect of physical gratings in the computational domain, thereby achieving fully digital decoding and recognition of the hidden card information. The entire method includes five core steps, forming a closed-loop collaborative structure of forward data flow driving and backward feedback optimization, as described in detail below.

[0015] Step S1: Multi-angle sequential illumination step. The technical purpose of this step is to provide light field information input with angular diversity for subsequent digital grating decoding. Unlike traditional solutions that use only uniform illumination from a single angle, this invention constructs a programmable multi-angle sequential illumination scheme based on a ring light source array. By illuminating the card surface from different incident angles, the microscopic features in the hidden information that have angular selectivity due to the coded grating structure produce differentiated reflection responses under different illumination conditions, providing sufficient angular dimension information for subsequent digital decoding.

[0016] Specifically, the ring light source array consists of N independently controllable LED light source units uniformly arranged along the circumference. In a preferred embodiment of the invention, N is 24. The light source units are evenly spaced on a circular support with a diameter of 12cm, forming a ring topology. Each LED light source unit contains four high-brightness white LED chips, with a luminous flux of 120lm per chip and an emission wavelength range covering the visible light band from 450nm to 650nm. The geometric center of the ring light source array is coaxially aligned with the optical axis of the image acquisition device. The card detection platform is located below the ring light source array, with a working distance set to 10cm. Under this geometric configuration, the incident angle formed by each light source unit relative to the normal direction of the card surface covers a range of -15° to +15°, and the angular interval between adjacent light source units is approximately 1.25°.

[0017] Preferably, the ring light source array achieves independent timing control of each light source unit through a light source driver controller. The light source driver controller uses an FPGA chip as the core control unit, with a timing accuracy of no less than 1μs, and supports preset illumination angle sequences. The light source units at corresponding positions are activated sequentially. In one embodiment of the invention, the illumination angle sequence adopts a linear scanning mode from -15° to +15°, i.e. ,in The duration of a single illumination is set to 20ms, and the switching interval between two adjacent illuminations does not exceed 2ms. Therefore, the sequential illumination process for all 24 angles takes approximately 528ms. In another embodiment of the present invention, the illumination angle sequence can also adopt a non-linear adaptive scanning mode. The sampling interval is increased in areas with strong angle sensitivity response, while the sampling interval is appropriately decreased in areas with weak response, so as to obtain the maximum amount of angle information within a limited number of samplings. In one embodiment of the present invention, the angle sequence determination process of the adaptive scanning mode is as follows: First, a fast pre-scan is performed with a large angle interval (e.g., 3.75°, corresponding to 8 angle positions). The information richness of each angle position is evaluated based on the gray-scale variance of the reflected image at each pre-scan angle position. An angle position with a large gray-scale variance indicates that the response of hidden information at that angle is more significant. Then, supplementary scanning is performed in the angle interval with a fine interval of 0.625° in the top 50% of the angle ranges with gray-scale variance, supplementing a total of about 16 angle positions, finally forming a mixed density illumination sequence containing 24 angle positions. This adaptive scanning strategy can improve the decoding signal-to-noise ratio by approximately 1.8 dB while maintaining the same total number of samples.

[0018] Furthermore, the light source driver controller also supports independent brightness adjustment for each LED light source unit, with a dynamic range of 10% to 100% of the rated luminous flux and an adjustment accuracy of 1%. This brightness adjustment capability enables the system to perform adaptive brightness compensation based on the reflective characteristics of different card materials, avoiding problems such as overexposure due to high-reflectivity areas on the card surface or insufficient signal due to low-reflectivity areas. In one embodiment of the present invention, before performing formal sequential illumination, the system first performs a rapid pre-scan at 50% of the rated brightness, automatically calculates the optimal brightness parameters for each angle position based on the grayscale histogram distribution of the pre-scanned image, and then performs formal sequential illumination acquisition using the optimal brightness parameters.

[0019] Step S2: Angle-Image Sequence Acquisition Step. This step receives the multi-angle sequential illumination output from Step S1 as input. Its technical purpose is to convert the optical reflection information of the card surface under each illumination angle into a structured digital image sequence, providing high-quality raw data for subsequent digital raster decoding.

[0020] The image acquisition device employs a CMOS image sensor with an area array. In a preferred embodiment of the invention, the sensor is an industrial-grade CMOS chip with a global shutter function, featuring an effective pixel count of 2048×2048, a pixel size of 3.45μm×3.45μm, a full-well capacity of 11000e-, and a readout noise of no more than 2.5e-. When used with a low-distortion industrial lens with a focal length of 25mm, the image acquisition device achieves a field of view of 8.4cm×8.4cm at a working distance of 10cm, meeting the full-frame acquisition requirements for standard cards (typically 6.3cm×8.8cm), with a spatial resolution of approximately 41μm / pixel.

[0021] During sequential illumination, strict synchronization control is achieved between the image acquisition device and the light source driver controller via hardware trigger signals. Specifically, whenever the light source driver controller activates a group of LED light source units, it synchronously sends a rising edge trigger pulse to the external trigger input port of the image acquisition device. Upon receiving the trigger signal, the image sensor immediately performs a global exposure acquisition. In one embodiment of the present invention, the exposure time of a single frame image is set to 15ms, slightly shorter than the 20ms continuous illumination time of the LED light source, to ensure that the exposure window falls entirely within the stable illumination range and to avoid image noise introduced by the transients of light source start-up and shutdown. The image acquisition device can achieve a frame rate of up to 45fps in sequential illumination mode, and the image bit depth is set to 14bit, providing quantization accuracy of 16384 gray levels.

[0022] After acquisition, all N frames of reflection images are organized into an angle-image sequence according to their corresponding illumination angle information. ,in Indicates the first lighting angle The reflected image is captured below. In one embodiment of the present invention, each frame of the image... The raw data size is 2048×2048×14bit ≈ 7.2MB, and the total data size of all 24 frames is approximately 172.8MB. To reduce the computational overhead of subsequent processing, the system performs preprocessing operations on the angle-image sequence, including three steps: dark current correction, flat field normalization, and lens distortion correction. Dark current correction eliminates the inherent dark current noise of the sensor by subtracting a pre-calibrated dark frame image; flat field normalization compensates for illumination non-uniformity and pixel gain differences by dividing by a pre-calibrated uniform white board response image; lens distortion correction eliminates barrel or pincushion distortion by inversely mapping the image coordinates using pre-calibrated distortion coefficients.

[0023] Preferably, the preprocessing also includes automatic location of the region of interest. The system uses a card edge detection algorithm to automatically identify the position and pose of the card in the image, extract the minimum bounding rectangle of the card region, and limit subsequent processing to this region to reduce the computational load on invalid background areas. In one embodiment of the invention, the card edge detection adopts a scheme combining edge extraction based on the Canny operator and line detection based on Hough transform, with the edge detection thresholds set to 0.3 times and 0.7 times the global grayscale mean of the image. The preprocessed angle-image sequence serves as the input data for digital raster decoding in step S3.

[0024] It is worth further elaborating that the data quality of the angle-image sequence directly determines the upper limit of the subsequent digital raster decoding effect. Therefore, this step designs a multi-layered monitoring mechanism for data quality assurance. First, after each frame is acquired, the system immediately calculates the sharpness evaluation index (based on the variance of the Laplacian operator). When the sharpness index is lower than a preset threshold, it is determined that the frame may have motion blur or defocus issues, and the system automatically discards the frame and re-acquires a frame under the same lighting conditions. In one embodiment of the present invention, the sharpness threshold is set to 60% of the average sharpness of the entire sequence, and in actual testing, the proportion of frames that need to be re-acquired does not exceed 2%. Second, after all frames are acquired, the system performs inter-frame consistency verification. By calculating the normalized mutual information between adjacent frames, it detects whether there are abnormal frames (such as occasional line noise from the sensor or brightness changes caused by light source flicker). Abnormal frames are repaired by weighted interpolation of adjacent frames. Through the above quality monitoring mechanism, the sharpness and consistency of each frame in the angle-image sequence can be maintained at a stable high level, laying a reliable data foundation for subsequent high-quality digital decoding.

[0025] Step S3: Digital Raster Decoding Step. This step is the core innovation of the entire method. Its technical purpose is to replace the optical filtering function of traditional physical gratings with purely digital computing methods, achieving contactless and grating-free decoding of the hidden information on the cards. This step receives the preprocessed angle-image sequence output from Step S2 as input and outputs the initial decoding result of the hidden information.

[0026] The core idea of ​​digital grating decoding lies in the fact that the selective transmission effect of a physical grating on hidden information is essentially a spatial frequency domain filtering operation. A physical grating arranges its transparent and opaque fringes at specific spatial frequencies and tilt angles. When superimposed on a printed surface containing hidden information, it only allows the image spatial frequency components that match the grating fringes' frequencies and angles to pass through, while suppressing mismatched components, thus highlighting the hidden information from the background pattern. Based on this physical principle, this invention constructs a digital grating transfer function in the computational domain, replacing the optical filtering process of the physical grating with mathematical operations.

[0027] Specifically, the digital grating transfer function In the frequency domain, it is defined as:

[0028] ,

[0029] in: and These are the horizontal and vertical frequency coordinates in the frequency domain space, respectively, with units of cycle / pixel; The spatial frequency of the digital grating is matched with the spatial frequency of the physical grating used to encode the hidden information on the card. In one embodiment of the present invention... The typical value range is 0.02 to 0.15 cycle / pixel, corresponding to a line density range of 50 to 300 lpi (lines per inch) for physical gratings; The orientation angle of the digital grating is consistent with the halftone angle of the coded grating, and its value ranges from 0° to 180°. This is the phase offset of the digital grating, used to precisely align the spatial phase of the hidden information, with a value ranging from 0 to... In practice, the optimal value is determined through a phase search algorithm; The window function is used to limit the effective range of the digital grating in the frequency domain, avoiding interference from noise frequency components far from the center frequency. In one embodiment of the present invention... A two-dimensional Gaussian window function is used, with its full width at half maximum (FWHM) set to the grating spatial frequency. 20%.

[0030] In the process of determining the parameters of the digital grating transfer function, the spatial frequency of the coded grating... and direction angle There are two approaches to obtaining the parameters. The first approach is the prior parameter input method, where the card's encoding parameters (including grating frequency, grating angle, and embedding position) are known information. These parameters are directly input into the system, which then constructs a precisely matching digital grating transfer function. The second approach is the automatic parameter estimation method, suitable for scenarios where the encoding parameters are unknown. The basic idea of ​​the automatic parameter estimation method is to perform a two-dimensional Fourier transform on typical frame images in the angle-image sequence, searching for significant frequency peaks in the spectrum other than the DC component. These peaks correspond to the spatial frequency and directional angle of the hidden information encoding grating. In one embodiment of this invention, the automatic parameter estimation algorithm first performs logarithmic enhancement and median filtering denoising on the spectrum, and then searches for the top 5 non-DC peaks in intensity in polar coordinates. The radial coordinates of these peaks are the candidate grating frequencies, and the angular coordinates are the candidate grating angles. The system sequentially tries each candidate parameter combination and selects the set with the best decoding effect as the final parameters.

[0031] Preferably, in the automatic parameter estimation method, the phase offset Determining the phase shift requires a more refined search process because the phase shift directly affects the spatial alignment accuracy between the digital grating and the encoded pattern of the hidden information on the card. In one embodiment of the present invention, the phase search algorithm employs a progressive refinement strategy: firstly, using... The step size is Eight candidate phase values ​​are coarsely searched within the range. A complete digital raster filter is performed on each candidate phase value, and the local contrast of the hidden information region in the resulting image is calculated. Then, with the optimal phase value as the center, [further analysis is performed]. Within the range A fine-grained search is performed with a step size, and the phase value with the largest local contrast is finally selected as the final value. The final value is determined. The total computational load of the above two-stage phase search process is approximately equivalent to 24 single-frame filtering operations, taking about 192ms at a resolution of 2048×2048 pixels. In one embodiment of the present invention, to further accelerate the phase search process, the image can be downsampled to 512×512 pixels to perform a coarse search, and then a fine search and final filtering can be performed only on the optimal candidate at the original resolution. This optimization strategy can reduce the phase search time to about 65ms.

[0032] Furthermore, the window function in the digital grating transfer function The design of the window function has a significant impact on decoding quality. An excessively narrow window function leads to a narrow filtering bandwidth, resulting in the loss of edge details hidden in the information; conversely, an excessively wide window function introduces excessive noise and background texture interference. In one embodiment of this invention, the optimal width of the window function is determined using a cross-validation method based on the decoding signal-to-noise ratio: the system in... Within the range of 10% to 30%, six different window widths were tried with a step size of 5%. The signal-to-noise ratio of the hidden information region was calculated for the decoding results under each window width, and the window width with the highest signal-to-noise ratio was selected as the final parameter.

[0033] The specific execution process of digital raster filtering is as follows: For each frame of the angle-image sequence... Perform a two-dimensional discrete Fourier transform to obtain its spectrum. The filtered spectrum is obtained by multiplying the spectrum point by point with the digital grating transfer function. Then, perform a two-dimensional inverse discrete Fourier transform on the filtered spectrum to obtain the decoded image of the frame. In one embodiment of the present invention, to improve computational efficiency, both the Fourier transform and the inverse transform are implemented using the Fast Fourier Transform algorithm. For a 2048×2048 pixel image, the transform-filter-inverse transform operation for a single frame takes approximately 8ms.

[0034] Preferably, to fully utilize the information redundancy provided by multi-angle illumination, this step also introduces a multi-frame fusion decoding mechanism. This involves decoding all N frames of images. The angle-sensitive response intensity is used as the weight for weighted averaging and fusion. Angle-sensitive response function Defined as the first Gray-scale contrast between the hidden information region and the background region in the frame-decoded image: ,in: For the first The average grayscale value of the hidden information region in the frame-decoded image. The average gray value of the background area. To prevent tiny positive numbers from being divided by zero, in one embodiment of the present invention Values The initial decoding result after fusion. Represented as: ,in: Angle - Total number of frames in the image sequence; For the first Frame angle sensitivity response weight; For the first The decoded image of the frame. Through this angle-sensitive weighted fusion strategy, angle positions with higher response intensity contribute more weight to the fusion result, while angle positions with weaker response naturally have lower weights. This results in a better initial decoding result in terms of signal-to-noise ratio than any single-frame decoded image. Tests in one embodiment of the invention show that multi-frame fusion decoding improves the signal-to-noise ratio by approximately 4.2 dB compared to the optimal single-frame decoding.

[0035] Step S4: Hidden Information Enhancement Step. This step receives the initial decoding result output from Step S3. The original angle-image sequence is used as input. The technical objective is to further improve the visibility and clarity of hidden information, especially for cards with low-contrast embeddings or slight surface wear. The enhancement process can improve the signal-to-noise ratio of the hidden content to a level that meets the requirements of subsequent semantic recognition. This step includes two cascaded sub-processes: multi-angle differential enhancement processing and frequency domain filtering processing.

[0036] The basic principle of multi-angle differential enhancement processing is that the background texture of the card surface (such as the original visible pattern, printed background, etc.) usually exhibits a relatively smooth and continuous change in reflectance intensity under different lighting angles. However, the embedded hidden information, due to the angle selectivity of the encoding grating structure, displays significant alternation of light and dark characteristics under different lighting angles. By utilizing this difference in angular response and calculating the difference between images at different angles, the gently changing background component can be effectively suppressed, while the drastically changing hidden information component can be preserved and highlighted.

[0037] Specifically, the mathematical expression for multi-angle difference enhancement is as follows. First, in the angle-image sequence... In the context of angle-sensitive response functions The frame with the largest value is taken as the reference frame, denoted as . The corresponding angle is Then, the weighted difference images between the remaining frames and the reference frame are calculated, resulting in multi-angle difference enhancement images. Defined as:

[0038] ,in: For the first The frame differential weighting coefficients, whose values ​​are determined by the angular deviation, are defined as follows: , The standard deviation parameter of the angle Gaussian kernel is given in one embodiment of the present invention. Setting it to 5° results in frames that deviate further from the reference frame angle receiving smaller differential weights, thus prioritizing the preservation of fine difference information between frames with angles close to the reference frame. Angle - Total number of frames in the image sequence; This indicates the absolute value operation. Multi-angle difference enhancement image. Then, normalization is performed to map its grayscale values ​​to the range of 0 to 255.

[0039] Preferably, before performing the difference operation, the system also performs sub-pixel-level image registration on each frame in the angle-image sequence to compensate for the slight image translation caused by changes in the illumination angle. The registration algorithm adopts a phase correlation method based on normalized cross-correlation, achieving a registration accuracy of 0.1 pixel, ensuring that the difference operation does not introduce false edges due to slight inter-frame displacement.

[0040] Frequency domain filtering is performed after multi-angle differential enhancement to further suppress residual noise components and mid-to-high frequency texture interference unrelated to hidden information in the enhanced image. A direction-sensitive bandpass filter is used for frequency domain filtering. Its transfer function is defined as: ,in: Frequency coordinates radial frequency, Its angular coordinates; The center frequency of the bandpass filter is set to the spatial frequency of the hidden information encoding grating. ; The bandwidth parameter in the frequency direction, in one embodiment of the present invention Set as 15%, meaning the 3dB bandwidth of the bandpass filter is approximately 30% of the center frequency; Select the angle for the direction, and set it as the direction angle of the encoding grating. ; An angular width parameter for direction selection, in one embodiment of the present invention Setting it to 3° gives the filter a selection range of approximately ±5° in the directional dimension.

[0041] The frequency domain filtering process is as follows: multi-angle differential enhancement image... The spectrum is obtained by performing a two-dimensional Fourier transform. The spectrum and direction-sensitive bandpass filter The enhanced hidden information image is obtained by multiplying point by point and then performing an inverse Fourier transform. In one embodiment of the present invention, the enhanced image is further subjected to contrast stretching and adaptive histogram equalization to further improve the visual recognition of the hidden pattern. After the cascaded processing of the above multi-angle differential enhancement and frequency domain filtering, the signal-to-noise ratio of the hidden information is significantly improved compared to the initial decoding result. It can further improve visibility by about 6.8dB, with a particularly significant improvement in the visibility of hidden content in low-contrast embedded scenes.

[0042] Preferably, this step also includes a quality pre-assessment of the enhancement result. The system calculates the enhanced hidden information image. The Weber contrast between the mid-foreground region (i.e., the hidden pattern region) and the background region is considered. When the Weber contrast is lower than the preset threshold of 0.15, it indicates that the current enhancement intensity is insufficient, and the system automatically adjusts the standard deviation of the differential weight kernel function. Reduced to 70% of the current value, and the bandwidth of the bandpass filter... After reducing the value to 80% of the current value, a stronger noise suppression and signal purification effect is achieved with a narrower filtering bandwidth, and then the enhancement process is repeated. In one embodiment of the present invention, this adaptive enhancement intensity adjustment only needs to be performed once on more than 99% of the test samples to meet the quality requirements, and two adjustments are required on about 0.8% of the difficult samples, with no more than three adjustment iterations in extreme cases. By introducing the above adaptive enhancement strategy, the enhancement processing effect of this step is more stable and consistent for card samples of different quality, avoiding the problem of over-enhancement or under-enhancement when facing cards with different contrast levels, which occurs with fixed parameter schemes. After the enhancement processing is completed, the output is an enhanced hidden information image. The data is then passed to step S5 for semantic recognition and comparison verification.

[0043] Step S5: Semantic recognition and comparison verification step. This step receives the enhanced hidden information image output from step S4. As input, the technical purpose is to automatically perform semantic understanding and authenticity judgment on the decoded and enhanced hidden content, and to feed back optimization signals to step S1 through comparison and verification results, forming a closed-loop control structure for the entire method.

[0044] Semantic recognition processing first involves processing the enhanced hidden information image. Adaptive binarization segmentation is performed to extract the hidden pattern from the background as a binary foreground region. Binarization segmentation employs a local adaptive thresholding algorithm, centered at each pixel in the image. A neighborhood window is used to calculate the weighted average of the pixel grayscale values ​​within the window as a local threshold. In one embodiment of the present invention... The value is 31 pixels, and the weighting method is Gaussian weighting. The foreground region after binarization and segmentation is denoised and hole-filled by morphological opening operation (erosion followed by dilation) and closing operation (dilation followed by erosion). The structuring element is a 3×3 circular template.

[0045] Subsequently, semantic feature vectors are extracted from the cleaned binary hidden pattern. In one embodiment of the present invention, the extraction of semantic feature vectors employs a multi-level feature fusion strategy, comprising three levels of feature components. The first level is global shape features, including a 7-dimensional vector of the Hu moment invariant of the hidden pattern, geometric parameters such as area ratio and compactness, totaling 10 dimensions. The second level is local texture features, calculated by extracting statistics such as energy, contrast, correlation, and uniformity of the gray-level co-occurrence matrix of the hidden pattern region, calculated in four directions (0°, 45°, 90°, 135°), totaling 16 dimensions. The third level is frequency domain energy distribution features, dividing the hidden pattern region into 4×4 sub-regions and calculating the energy concentration index of the two-dimensional Fourier transform spectrum of each sub-region, totaling 16 dimensions. The feature components of the three levels are concatenated to form a 42-dimensional semantic feature vector. .

[0046] The comparison and verification process will extract semantic feature vectors. The similarity is calculated between the feature vectors and the pre-stored standard hidden information template library. The standard template library stores the standard hidden information feature vectors of various legal cards. ,in This represents the capacity of the template library. The similarity comparison uses cosine similarity as a metric, defined as: ,in: This represents the semantic feature vector of the card to be detected. For the first in the template library One standard feature vector; Let be the Euclidean norm of the vector. When Greater than or equal to the preset judgment threshold When, the hidden information is determined to be true and complete; when Less than At that time, it is determined that the hidden information is abnormal. In one embodiment of the present invention, The default value is 0.90.

[0047] Preferably, in addition to cosine similarity, the system also introduces a quantitative evaluation index for the integrity of hidden information. This is used for a more refined quantitative evaluation of the clarity and anti-counterfeiting effectiveness of hidden patterns. ,in: The structural clarity score is obtained by calculating the ratio of the Sobel gradient energy of the enhanced hidden pattern edge to the gradient energy of the ideal template edge, with a value ranging from 0 to 1. The content completeness score is obtained by calculating the ratio of the area of ​​the connected region of the enhanced hidden pattern to the area of ​​the connected region of the standard template, with a value ranging from 0 to 1. Decoding confidence is defined as the initial decoding result. The normalized signal-to-noise ratio value of the hidden information region, ranging from 0 to 1; , and The weighting coefficients for the three indicators are 0.4, 0.35, and 0.25 respectively in one embodiment of the present invention, and their sum is 1. When When the decoding quality is excellent, it is judged to be good; when When, it is judged as qualified; when If the decoding quality is deemed unqualified, a re-inspection will be triggered.

[0048] The key innovation of this step lies in establishing a closed-loop feedback mechanism from semantic recognition results to multi-angle sequential illumination. When comparing similarity... Below the judgment threshold or integrity assessment indicators When the quality falls below the acceptable level, the system does not immediately issue a negative judgment. Instead, it first analyzes the reasons for the poor decoding quality. Specifically, the system calculates the sharpness distribution map of each local area in the enhanced hidden pattern, identifies weak areas with sharpness below 50% of the global average, and infers the optimal illumination angle range corresponding to the weak area based on its spatial location on the card. In the next round of detection, the illumination angle sequence in step S1 will increase the sampling density within the angle range corresponding to the weak area. For example, the angle interval within this range will be reduced from the default 1.25° to 0.5°, while the brightness of the LED light source within this angle range will be appropriately increased to 120% of the rated value. Through this closed-loop feedback optimization mechanism, the system can improve the decoding quality to an acceptable level after a maximum of three rounds of iteration detection, with an iteration convergence rate of over 97% in actual tests.

[0049] Furthermore, the output of this step also includes structured detection report information, which includes card number, detection timestamp, decoded image, and similarity comparison. Integrity score The final authenticity determination result facilitates subsequent traceability and data analysis. In one embodiment of the present invention, the entire process from sequential illumination initiation to output of detection result for a single card takes no more than 1 second. Specifically, illumination acquisition in steps S1 and S2 takes approximately 530ms, digital raster decoding in step S3 takes approximately 200ms, enhancement processing in step S4 takes approximately 150ms, and identification verification in step S5 takes approximately 80ms. The remainder is system scheduling overhead. This processing speed can meet the batch detection throughput requirement of more than 3600 cards per hour.

[0050] like Figure 2 As shown, the visual recognition system for hidden card information based on one-click decoding provided in this embodiment of the invention includes a multi-angle sequential illumination module, an angle-image sequence acquisition module, a digital grating decoding module, a hidden information enhancement module, and a semantic recognition and comparison verification module. The modules are interconnected through a data bus and a control signal bus, forming a hardware and software collaborative working architecture that corresponds one-to-one with each step of the above method embodiment.

[0051] The multi-angle sequential lighting module corresponds to step S1 in the method embodiment. Its hardware components include a ring light source array and a light source driver controller. The physical structure of the ring light source array, as described in the method embodiment, consists of 24 groups of LED light source units evenly distributed along a 12cm diameter circular support. The light source driver controller is implemented based on an FPGA chip, and its internal firmware stores various preset lighting angle sequence schemes, including a linear uniform scanning scheme, a center-encrypted scanning scheme, and an adaptive optimization scanning scheme. This module receives feedback optimization parameters output by the semantic recognition and comparison verification module through a control signal bus, dynamically adjusting the sampling density and brightness parameters of the lighting angle sequence to achieve closed-loop adaptive control. In one embodiment of the invention, the light source driver controller also integrates a temperature monitoring function. When the junction temperature of the LED light source array exceeds 65°C, it automatically reduces the driving current to protect device lifespan and ensure the stability of the system under long-term continuous operation conditions.

[0052] The angle-image sequence acquisition module corresponds to step S2 in the method embodiment. Its hardware components include an area array CMOS image sensor, a low-distortion industrial lens, and an image acquisition control circuit. The image acquisition control circuit maintains strict synchronization with the light source drive controller through a hardware trigger interface, ensuring that one frame of image acquisition is precisely triggered each time illumination is activated. The software part of this module is responsible for image preprocessing operations, including dark current correction, flat field normalization, distortion correction, and automatic localization of the region of interest on the card. These preprocessing parameters are predetermined through a calibration process and stored in non-volatile memory before the system leaves the factory. Preferably, the module also includes a high-speed cache memory with a capacity of not less than 512MB for temporarily storing the original angle-image sequence data of the current detection batch, supporting the on-demand retrieval of the original image data after detection is completed.

[0053] The digital raster decoding module corresponds to step S3 in the method embodiment, and its core is a digital raster decoding algorithm engine running on a high-performance embedded processor. In one embodiment of the present invention, the embedded processor is an edge computing platform with GPU acceleration capabilities, a main frequency of not less than 1.5 GHz, and equipped with not less than 4 GB of DDR4 RAM. The digital raster decoding algorithm engine implements the digital raster transfer function. The module handles all core computational processes, including card construction, fast Fourier transform filtering, and multi-frame fusion decoding. It supports two operating modes: prior parameter mode and automatic estimation mode, allowing users to switch flexibly according to their application scenarios. In prior parameter mode, the system retrieves grating parameters matching the current card model from its built-in card encoding parameter database; in automatic estimation mode, the system automatically infers grating parameters based on spectral analysis. The module outputs the initial decoding results to the hidden information enhancement module.

[0054] The hidden information enhancement module corresponds to step S4 in the method embodiment. It is implemented in software on the same embedded processor as the digital grating decoding module and includes two cascaded functional subunits: a multi-angle differential enhancement processing unit and a frequency domain filtering processing unit. The algorithm implementation of the multi-angle differential enhancement processing unit is as described in the differential enhancement formula in the method embodiment, and the frequency domain filtering processing unit implements a direction-sensitive bandpass filter. Frequency domain filtering operations are performed. Preferably, the hidden information enhancement module also integrates an adaptive parameter adjustment function, which can automatically adjust the differential weight parameters according to the signal-to-noise ratio level of the initial decoding result. Bandwidth parameters of bandpass filters To accommodate the enhanced processing needs of cards of different quality levels. When the initial decoding signal-to-noise ratio is below 15dB, the system automatically... Narrowed from the default 5° to 3°, and from The noise level was narrowed from 15% to 10% to achieve a stronger noise suppression effect.

[0055] The semantic recognition and comparison verification module corresponds to step S5 in the method embodiment, and includes a semantic feature extraction unit, a template comparison unit, and a closed-loop feedback control unit. The semantic feature extraction unit extracts 42-dimensional feature vectors, including Hu moment invariants, gray-level co-occurrence matrix statistics, and frequency domain energy distribution features. The template comparison unit has a built-in standard template feature database and supports cosine similarity calculation and integrity evaluation metrics. Quantitative evaluation. The closed-loop feedback control unit generates lighting angle optimization parameters based on the comparison and verification results, and feeds them back to the multi-angle sequential lighting module through the control signal bus to achieve adaptive iterative optimization of the detection parameters. In one embodiment of the present invention, the standard template feature database supports online updates and expansion via USB interface or network interface, and the template library capacity supports feature storage for no less than 10,000 card models.

[0056] All the above modules are physically integrated into a compact testing device. The overall dimensions of the device do not exceed 30cm × 25cm × 20cm, the weight does not exceed 5kg, the power supply voltage is 12V DC, and the rated power does not exceed 60W. The device is equipped with a 7-inch touch screen as the human-machine interface, supporting one-button start-up. Users only need to place the card to be tested on the testing platform and press the start button. The system automatically executes the entire process of illumination-acquisition-decoding-enhancement-recognition and outputs the test results on the display screen, achieving a truly one-button decoding experience. Preferably, the system also supports data communication with the host computer management system via Ethernet or wireless network interface, supporting real-time uploading of test results, remote monitoring, and statistical report generation, meeting the needs of industrial production line integration and deployment. The embedded operating system built into the testing device adopts the Linux kernel, supporting multi-task parallel scheduling. It can simultaneously handle the testing process of the current card and the data upload task of the previous card's test result, without affecting the testing cycle due to network communication.

[0057] To fully verify the effectiveness and superiority of the technical solution of this invention, this embodiment constructed a test card set containing 2000 samples across 5 different brands and different methods of embedding hidden information. The test cards covered various combinations of encoding parameters, with grating frequencies ranging from 80 lpi to 200 lpi and grating angles from 0° to 75°. The test environment was a standard darkroom with ambient light levels below 5 lux, a temperature controlled at 23±2°C, and humidity controlled at 45±5%RH.

[0058] Regarding digital decoding success rate, the initial decoding success rate for all 2000 test cards was 99.5% (1990 / 2000). Analysis of the 10 failed decoding samples revealed that they were all extreme cases of severe physical damage (such as large-area creases or stains covering more than 30% of the hidden area). In terms of angle support range, the system successfully decoded hidden information within a lighting angle range of -15° to +15°. Within the core angle range of -10° to +10°, the decoding signal-to-noise ratio was generally higher than 25dB, and the decoding signal-to-noise ratio within the edge angle ranges of -15° to -10° and +10° to +15° was no less than 18dB.

[0059] In terms of decoding time, the average time for a single card from detection start to result output is 0.96 seconds, with a maximum time not exceeding 1.2 seconds (corresponding to complex scenarios requiring two rounds of iterative optimization), which is far lower than the 30 to 60 seconds of operation time typically required for manual decoding using traditional physical gratings. Based on a detection cycle of approximately 1 second per card, the system can achieve a batch detection throughput of over 3600 cards per hour, which is approximately 60 to 120 times faster than traditional manual methods.

[0060] Regarding the accuracy of hidden content recognition, the cosine similarity threshold was used. As a criterion, the system achieved a recognition accuracy of 98.0% (1950 / 1990) on 1990 successfully decoded cards, with a true positive rate of 98.5% and a false positive rate of 0.3%. Further, integrity assessment indicators... The average score was 0.87, indicating that the decoded hidden pattern achieved a high level in all three dimensions: structural clarity, content completeness, and decoding confidence.

[0061] To verify the effectiveness of the closed-loop feedback optimization mechanism, this embodiment performs a test on the initial detection. 128 card samples with scores below 0.80 underwent a second iteration of testing. After one round of closed-loop feedback optimization, 117 of these samples... The score improved to above 0.80, with an optimization success rate of 91.4%. After two rounds of optimization, 8 out of the remaining 11 images reached the qualified level, resulting in a cumulative optimization success rate of 97.7%. The above test results fully verify that the technical solution of this invention is significantly superior to the traditional physical grating decoding method in terms of digital decoding capability, detection efficiency, and recognition accuracy.

[0062] To further illustrate the performance contribution of each technical aspect of this invention, an ablation experiment was conducted in this embodiment. In the ablation experiment, the detection performance of the following configurations was tested: using only single-angle illumination with digital grating decoding (without multi-angle fusion), using multi-angle illumination with digital grating decoding but without hidden information enhancement, using the complete process but disabling closed-loop feedback optimization, and the complete process of this invention. The results show that when using only single-angle illumination, the decoding success rate drops to 82.3%, and the recognition accuracy drops to 79.6%, indicating that the information redundancy provided by multi-angle illumination makes a crucial contribution to decoding quality. Without hidden information enhancement, the decoding success rate remains at 98.7% (because the decoding step has already completed the main information extraction), but the recognition accuracy drops to 91.2%, indicating that the enhancement process has a significant effect on extracting reliable semantic information from low-contrast decoding results. Disabling closed-loop feedback optimization has no impact on samples that have already been decoded with high quality in the first detection, but the final recognition accuracy for edge samples (approximately 6.4% of the total samples) decreases by about 3.5 percentage points, indicating that the closed-loop feedback mechanism plays an irreplaceable role in dealing with difficult samples.

[0063] Furthermore, the robustness of the system was tested in this embodiment. The detection performance was tested under three artificial damage conditions: simulated abrasion (lightly abrading the card surface 10 times with 320-grit sandpaper), simulated oil stains (adding 0.05 mL of cooking oil and wiping), and simulated bending (bending once diagonally and then flattening). Under the light abrasion condition, the decoding success rate was 97.8%, and the recognition accuracy rate was 95.2%; under the oil stain condition, the decoding success rate was 96.5%, and the recognition accuracy rate was 93.7%; under the bending condition, the decoding success rate was 98.2%, and the recognition accuracy rate was 96.1%. These robustness test results demonstrate that the computational imaging decoding scheme of this invention has good tolerance to common physical damage to cards, meeting the anti-counterfeiting detection requirements in actual circulation environments.

[0064] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for visually recognizing hidden card information based on one-click decoding, characterized in that, Includes the following steps: Multi-angle sequential illumination step: The surface of the card to be tested is illuminated with programmable angles through a ring light source array. The ring light source array contains multiple sets of independent controllable LED light source units evenly distributed along the circumference. The light source units at the corresponding positions are activated sequentially according to the preset illumination angle sequence, so that the illumination beam illuminates the card surface at different incident angles, forming a sequential illumination process covering the preset angle range. Angle-image sequence acquisition steps: Under each illumination angle, the reflected image of the card surface is synchronously acquired by the image acquisition device. The reflected images acquired under all illumination angles are organized into an angle-image sequence in the order of illumination, where each frame of the image forms an associated mapping relationship with the corresponding illumination angle information. Digital raster decoding steps: Perform digital raster decoding processing on the angle-image sequence. By constructing a digital raster transfer function that matches the encoding parameters of the hidden information on the card, apply digital raster filtering operation to each frame of the angle-image sequence to simulate the selective transmission effect of the physical raster on the hidden information at a specific viewing angle, and extract the initial decoding results of the hidden graphics and image information embedded in the card. Hidden information enhancement steps: Perform multi-angle differential enhancement processing and frequency domain filtering processing on the initial decoding result. The multi-angle differential enhancement processing suppresses the background texture of the card surface and highlights the angle-sensitive features of the hidden information by calculating the weighted difference between the images under different lighting angles. The frequency domain filtering processing further improves the contrast and clarity of the hidden content by constructing a bandpass filter that matches the spatial frequency characteristics of the hidden information, thus obtaining the enhanced hidden information image. Semantic recognition and comparison verification steps: Perform semantic recognition processing on the enhanced hidden information image, extract the semantic feature vector of the hidden content, and compare the semantic feature vector with the pre-stored standard hidden information template. Determine the authenticity and completeness of the hidden information based on the comparison similarity.

2. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, In the multi-angle sequential illumination step, the ring light source array contains N groups of LED light source units, where N is an integer between 8 and 36. The illumination angle coverage range of each light source unit is -15° to +15°, the angle interval between adjacent light source units is 0.8° to 3.75°, the wavelength range of the illumination beam is 450nm to 650nm, and the duration of a single illumination is 5ms to 50ms.

3. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, In the angle-image sequence acquisition step, the acquisition frame rate of the image acquisition device is not less than 200fps, the image resolution is not less than 1024×1024 pixels, the working distance between the image acquisition device and the card surface is 5cm to 20cm, and the bit depth of the reflected image is 12bit to 16bit.

4. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, In the semantic recognition and comparison verification step, the threshold for judging the comparison similarity is 0.85 to 0.

95. When the comparison similarity is greater than or equal to the threshold, the hidden information is judged to be true and complete. When the comparison similarity is less than the threshold, the hidden information is judged to be abnormal and a re-inspection process is triggered.

5. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, In the digital grating decoding step, the construction process of the digital grating transfer function includes: constructing a periodic transfer function with corresponding spatial frequency and direction selectivity in the frequency domain space based on the coded grating frequency and coded grating angle of the card hidden information; performing point-by-point multiplication operation on the transfer function and the spectrum of each frame image in the angle-image sequence and then performing an inverse transformation to obtain the decoded image after digital grating filtering.

6. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, The specific process of multi-angle differential enhancement in the hidden information enhancement step includes: taking the image with the largest response intensity in the angle-image sequence as the reference frame, calculating the weighted differential images between the remaining frames and the reference frame, wherein the differential weight of each frame is determined by the angle sensitivity response function, and then normalizing and accumulating all weighted differential images to obtain the multi-angle differential enhancement image.

7. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, In the hidden information enhancement step, the frequency domain filtering process uses a direction-sensitive bandpass filter. The center frequency of the bandpass filter is set to the spatial frequency of the hidden information encoding grating, the bandwidth is set to 10% to 30% of the center frequency, and the direction selection angle is set to within ±5° of the encoding grating angle.

8. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, The ring light source array also includes a light source driver controller, which is connected to each LED light source unit through an independent channel. It supports controlling the opening, closing and brightness adjustment of each light source unit according to a preset timing sequence. The timing accuracy of the light source driver controller is not less than 1μs.

9. The method for visually recognizing hidden card information based on one-click decoding according to claim 1, characterized in that, The comparison and verification results are fed back to the multi-angle sequential illumination step to adaptively optimize the illumination angle sequence for the next round of detection. The adaptive optimization process of feeding back the comparison and verification results to the multi-angle sequential illumination step includes: when the comparison similarity is lower than the judgment threshold, the angle distribution of the weak area of ​​hidden information decoding is determined according to the decoding quality evaluation index, and the illumination density and illumination times of the corresponding angle range of the weak area are increased in the next round of detection.

10. A card hidden information visual recognition system based on one-click decoding, used to implement the card hidden information visual recognition method based on one-click decoding as described in any one of claims 1-9, characterized in that, include: The multi-angle sequential illumination module is used to perform programmable angle sequential illumination on the surface of the card to be inspected through a ring light source array. According to the preset illumination angle sequence, the light source units at the corresponding positions are activated in sequence to form a sequential illumination process covering the preset angle range. Angle-image sequence acquisition module is used to synchronously acquire the reflected image of the card surface under each illumination angle, and organize the reflected images acquired under all illumination angles into an angle-image sequence in the order of illumination. The digital grating decoding module is used to extract the initial decoding result of the hidden information by constructing a digital grating transfer function that matches the encoding parameters of the hidden information on the card, applying digital grating filtering operation to the angle-image sequence, simulating the selective transmission effect of the physical grating; The hidden information enhancement module is used to perform multi-angle differential enhancement processing and frequency domain filtering processing on the initial decoding result, suppressing background texture and improving the contrast and clarity of the hidden content; The semantic recognition and comparison verification module is used to perform semantic recognition on the enhanced hidden information image and compare it with the standard template to determine the authenticity and completeness of the hidden information. The verification results are then fed back to the multi-angle sequential lighting module to adaptively optimize the lighting angle sequence.

Images