Wine package digital watermark embedding identification system and method

Abstract

The invention is suitable for the technical field of packaging engineering and digital anti-counterfeiting, and provides a wine packaging digital watermark embedding identification system and method, the wine packaging digital watermark embedding identification system comprises a multi-layer structure from a base material to a protective layer, and an improved DCT watermark embedding algorithm, an ultraviolet laser etching microstructure technology and a high-speed online visual identification unit are integrated. According to the method, through algorithm embedding, laser physical writing and camera rapid extraction verification, extremely low cost, high anti-copying capacity and single material environment-friendly recovery of package anti-counterfeiting are achieved. The problems of high cost, easy copying and complex process in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the fields of packaging engineering and digital anti-counterfeiting technology, specifically relating to a digital watermark embedding and recognition system and method for wine packaging, and in particular a low-cost, highly concealed, and strongly anti-copying digital watermark solution based on an improved discrete cosine transform (DCT) algorithm, combined with laser etching microstructure manufacturing and high-speed online visual recognition technology. Background Technology

[0002] In the field of anti-counterfeiting for alcoholic beverages, companies commonly use laser-etched aluminum foil labels, RFID fragile tags, or exposed QR codes for identification and consumer verification. Laser-etched aluminum foil labels rely on a visible laser pattern for basic identification, RFID tags store traceability information via radio frequency chips, and exposed QR codes rely on a backend database for online queries. Laser-etched labels are made by laminating PET aluminum foil, die-cutting, and inkjet printing, with a cost of approximately 0.04–0.06 yuan per label; RFID tags require embedding a chip and antenna on coated paper before encapsulation, with a cost of 0.25–0.35 yuan per label; exposed QR codes are directly inkjet printed, offering the lowest cost but are easily copied by photographing. 1. Conflict between cost and environmental protection: Laser aluminum film labels require secondary lamination with PET laser film, the material cannot be recycled alone, and aluminum prices fluctuate greatly; RFID chips rely on imported wafers, with long delivery cycles and high prices; 2. High risk of duplication and recycling: Exposed QR codes can be screenshotted with a single scan, and genuine codes can be reused on counterfeit goods; Although laser-marked patterns are complex, they can still be electro-engraved and restored after high-definition scanning. 3. Lengthy process chain: Laser labeling requires three processes: lamination, die-cutting, and inkjet printing; RFID requires patching, packaging, and coding. The equipment investment is large and the changeover time is long, making it difficult to meet the flexible demand of the liquor industry for "small batches and multiple SKUs". While some researchers have attempted to embed digital watermarks into packaging images, most methods employ global DCT or spatial LSB, which have weak resistance to print-photograph-reprint (PRP) and generally have error rates exceeding 15%. Therefore, the market urgently needs a digital watermarking solution for wine packaging that offers near-raw watermark costs, compatibility with existing color printing processes, and strong resistance to duplication. Summary of the Invention

[0003] The purpose of this invention is to provide a digital watermark embedding identification system for wine packaging, which aims to solve the technical problems of high cost, complex process, and poor environmental performance of laser labels and RFID tags, as well as the easy copying and insufficient robustness of QR codes and traditional digital watermarks in the prior art.

[0004] This invention is implemented as follows: a digital watermark embedding and recognition system for wine packaging, comprising, from top to bottom: Substrate layer; A base ink layer is coated on the substrate layer to carry printed graphic information; A laser-etched microstructure layer is formed on the base ink layer, containing a microscopic physical structure corresponding to the embedded digital watermark information; A surface protective varnish layer is applied over the laser-etched microstructure layer to protect the microscopic physical structure. An online imaging layer is used to acquire images of the packaging area containing the laser-etched microstructure layer; The system further includes a processing unit, which is used to embed the digital watermark information to be embedded into the packaging image through a discrete cosine transform algorithm, control the laser device to form the laser-etched microstructure layer on the surface protective varnish layer or a designated layer according to the embedded image data, and control the online imaging layer to acquire images and extract the digital watermark information from the acquired images for verification.

[0005] A further technical solution of the present invention is as follows: the laser-etched microstructure layer is formed by laser etching, with a depth of 0.4-0.6µm, and physically writes 768 bits of information, including a 512-bit digital payload; the laser-etched microstructure layer is etched using a diode-pumped all-solid-state Nd:YVO4 laser, and the laser is a TEM. 00 The fundamental mode has a wavelength of 355nm, an average power of 2.5-3.5W, a pulse frequency of 50-70kHz, a single pulse energy of 50µJ, a pulse width of 15ns, a bidirectional grid scanning path, a line spacing of 6µm, and a scanning speed of 120m / min, which is synchronized with the printing production line; the microscopic physical structure is an array of pits formed in the region corresponding to each image block, with a pit depth of 0.2-0.3µm and a diameter of 3-5µm; In the laser-etched microstructure layer, each 8×8 pixel DCT block corresponds to a 120µm×120µm laser "cell". When the DCT coefficient needs to be +α, a 2×2 array of pits is etched in the cell, with a pit depth of 0.25µm and a diameter of 4µm. When –α is needed, it is skipped. Paired coefficients (2,3) and (3,2) share the same cell, and the pit array is rotated 90° to distinguish them.

[0006] A further technical solution of the present invention is: the discrete cosine transform algorithm is a block DCT algorithm, which divides the image into multiple 8×8 pixel blocks. In each block, the two low-frequency coefficient positions with the smallest quantization step size in the DCT domain are selected for pairwise embedding of watermark bits. The embedding strength is adaptively adjusted according to the standard deviation of the pixels in the block. The digital watermark embedding and recognition system for wine packaging uses an improved DCT algorithm to embed watermarks. It embeds the same 1-bit information in pairs only in the low-to-medium frequency coefficients (2,3) and (3,2) with the smallest JPEG quantization step size. The embedding amount α is linearly scaled with the local standard deviation σ of the image. The embedding order is written after global shuffling. The extraction is performed by voting. The embedding rule of the block-based DCT algorithm is as follows: for the DCT coefficients J of 8×8 blocks i and j, when writing bits w∈{0,1}, J2,3←J2,3+α (2w 1), J3,2←J3,2+α (2w 1); Extraction rules are as follows =½[sgn (J2,3+J3,2)+1]; The base value of the embedding quantity α is 2, and it is linearly mapped to [0.8, 2] by the block standard deviation σ, with the mapping formula α = 0.8 + 1.2. σ / 25, where σ is the standard deviation of an 8×8 pixel block.

[0007] A further technical solution of the present invention is: the watermark extraction algorithm executed by the processing unit includes: performing block DCT transformation on the image acquired by the online shooting layer, accumulating the symbols of the two specified low-frequency coefficients and performing voting judgment to recover the watermark bit sequence, comparing the extracted sequence with the reference sequence in the cloud or locally, and judging the authenticity based on the Hamming distance; The image preprocessing of the online shooting layer includes: encoder triggering side light for 20µs exposure to obtain the Cb channel, perspective correction through Hough edge creasing lines, and smoothing out the vignetting with a 31×31 white cap. The watermark extraction algorithm in the online shooting layer is ARM-Neon parallel 8×8 DCT operation, accumulating coefficients (2,3)(3,2) symbol voting to obtain bits, looking up the table and reverse shuffling 4096 bits, with a total extraction time ≤16ms. The watermark matching mechanism is as follows: the local 512-bit extracted sequence is encrypted with AES-CTR and sent to the cloud. It is XORed with the reference sequence to obtain the Hamming weight HD. HD≤18 is considered true, 19–30 is a second image is taken, and >30 is directly rejected. The result is written back to MySQ as an HMAC-SHA256 digest.

[0008] A further technical solution of the present invention is: the digital watermark embedding and recognition system for wine packaging further includes a synchronization and calibration unit for achieving precise alignment between laser etching and printed graphics; the synchronization and calibration unit includes an encoder installed on the drive side of the printing press and a photoelectric sensor for detecting printing overprint marks, the signal of the encoder is used to trigger the start and cycle of laser etching, and the mark position deviation detected by the photoelectric sensor is used to provide real-time feedback to adjust the coordinate origin of the laser scan.

[0009] A further technical solution of the present invention is: the digital watermark embedding and recognition system for wine packaging maintains alignment accuracy through closed-loop control logic. The closed-loop control logic includes: periodically calculating the offset between the actual etching position and the theoretical position by acquiring images through the online imaging layer. If the offset exceeds a threshold, the driving parameters of the laser scanning mirror or the encoder origin compensation value are automatically adjusted.

[0010] A further technical solution of the present invention is: the refractive index of the surface protective varnish layer is 1.50-1.52, which forms a refractive index difference with the air in the pits of the laser-etched microstructure layer, so that a shadow contrast of more than 8 gray levels is generated under side lighting.

[0011] A further technical solution of the present invention is: the substrate layer is a single-material white cardboard or biodegradable polylactic acid film, the base ink layer is a CMYK four-color flexographic ink layer with a thickness of 1-3μm, the surface protective varnish layer has a thickness of 1-2μm and is a UV-curable varnish, and the wine packaging digital watermark embedded identification system as a whole constitutes a packaging structure that can be recycled from a single material.

[0012] Another objective of this invention is to provide a method for embedding and recognizing digital watermarks on wine packaging, applied to the aforementioned wine packaging digital watermark embedding and recognition system, the method comprising: S1. The digital watermark information is embedded into the preset image of the wine packaging through the improved discrete cosine transform algorithm to generate watermarked image data. S2. Based on the watermarked image data, control the laser device to etch a laser-etched microstructure layer corresponding to the change of DCT coefficient on a designated layer of the printed packaging material; S3. Acquire images of the area containing the laser-etched microstructure layer on the packaging of the finished product using an online imaging device; S4. Extract digital watermark information from the acquired image, match and verify it with the original embedded information, and output the authenticity judgment result.

[0013] A further technical solution of the present invention is: in step S1, the embedding intensity is adaptively adjusted according to the local texture complexity of the image block, and the adjustment formula is: α = 0.8 + k·σ / 25, where α is the embedding intensity, σ is the standard deviation of pixels in the block, and k is a constant; The laser etching parameters in step S2 include: wavelength 355nm, average power 3.0W±0.05W, pulse frequency 50-70kHz, scanning speed synchronized with the production line speed of 120m / min, and each 8×8 pixel DCT block mapped to a 120µm×120µm physical etching cell. In step S4, the algorithm extracts and calculates the DCT coefficients of each block in parallel. The symbols of two specified low-frequency coefficients are accumulated and voted to obtain the watermark bits. The Hamming distance between the extracted sequence and the reference sequence is calculated to determine the authenticity, and the authenticity threshold is ≤18.

[0014] The beneficial effects of this invention are: it eliminates expensive consumables such as aluminum foil, laser film, and RFID chips, reducing the cost of a single watermark tag to approximately 0.003 yuan, only 6-8% of the cost of traditional laser tags and about 1% of the cost of RFID tags, significantly reducing anti-counterfeiting costs for enterprises; by embedding low-to-medium frequency DCT coefficients in pairs, adaptive intensity adjustment, and laser etching to form a difficult-to-replicate physical microstructure, it significantly improves the ability to resist print-photograph-reprint attacks. Experiments show that after 10 PRP cycles, the bit error rate can be stably below 3%, a reduction of more than 80% compared to traditional global DCT schemes. The online identification and decoding success rate can reach over 99.7%, providing stronger anti-counterfeiting reliability. The system can be directly integrated into the end of existing printing presses, using laser etching to physically write watermarks in a single process, eliminating multiple steps such as lamination, die-cutting, and separate coding. This adapts to the flexible production needs of the liquor industry, which requires small batches and multiple SKUs, reducing energy consumption and labor costs. The packaging structure tends to use a single material, simplifying the recycling process, improving the purity and value of recycled materials, and reducing environmental pollution. Consumers can quickly verify authenticity via the cloud by taking a photo with a regular smartphone, providing a user-friendly experience. High-speed online full inspection or sampling inspection can be achieved on the production line to ensure the quality of products leaving the factory. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of the digital watermark embedding and recognition system for wine packaging provided in an embodiment of the present invention.

[0016] Figure 2 This is a flowchart of the packaging digital watermark embedding and recognition method provided in an embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram comparing the original image and the watermark provided in the embodiments of the present invention.

[0018] Figure label: Detailed Implementation Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0019] like Figure 1 As shown, the present invention provides a digital watermark embedding and recognition system for wine packaging. Its structure, from top to bottom, includes a substrate layer, a base ink layer, a laser-etched microstructure layer, a surface protective varnish layer, and also includes an online imaging layer and a processing unit.

[0020] The substrate layer is a printed substrate, using a single material of white cardboard or biodegradable polylactic acid film, ensuring that the packaging can be recycled from a single material.

[0021] The base ink layer is coated on the substrate layer, with a thickness of 1-3μm, and consists of conventional four-color flexographic inks, used to carry CMYK graphic information.

[0022] Laser etching was used to form a microstructure layer on top of the base ink layer, with a depth of 0.4–0.6 µm. 768 bits of information, including a 512-bit digital payload, were physically written. The etching was performed using a diode-pumped all-solid-state Nd:YVO4 laser, which was a TEM laser. 00 The fundamental mode has a wavelength of 355 nm, an average power of 2.5-3.5 W (preferably 3.0 W ± 0.05 W), a power density of 0.95 MW / cm², a pulse frequency of 50-70 kHz (preferably 60 kHz), a single pulse energy of 50 µJ, a pulse width of 15 ns, a peak power of 3.3 kW, and an overlap of 30% between adjacent pulses. The scanning path is a bidirectional grid scan with a line spacing of 6 µm (= spot diameter 1 / e²), a scanning speed of 120 m / min (= 2 m / s), corresponding to 3.3 mm / line negative pressure suction for stabilization and confocal ranging closed-loop ± 2 µm. The microscopic physical structure is a pit array, with each 8×8 pixel DCT block corresponding to a 120µm×120µm laser "cell". When the DCT coefficient needs to be +α, a 2×2 array of pits is engraved in the cell, with a pit depth of 0.2-0.3µm, preferably 0.25µm, and a diameter of 3-5µm, preferably 4µm, resulting in a 0.4% decrease in macroscopic reflectivity. When –α is needed, the pits are skipped. Paired coefficients (2,3) and (3,2) share the same cell, and the pit array is rotated 90° to distinguish them.

[0023] A surface protective varnish layer, 1-2 μm thick, is applied over the laser-etched microstructure layer. It is a UV-curable varnish with a refractive index of 1.50-1.52. This varnish creates a refractive index difference with the air in the pit, resulting in a shadow contrast of more than 8 gray levels under side lighting. This is used to protect the DCT coefficients from wear.

[0024] The online imaging layer includes a ≥300dpi CCD linear array and a side-lighting unit, used to acquire images of packaging areas containing laser-etched microstructure layers in real time and upload them to the database. The image preprocessing process involves encoder-triggered side-lighting for 20µs exposure to directly capture the Cb channel, using Hough edge creasing lines for perspective correction, and then smoothing out vignetting with a 31×31 white cap. The entire 720×1280 image is completed in 4ms with a 18% contrast improvement.

[0025] The processing unit is used to embed the digital watermark information to be embedded into the packaging image through an improved discrete cosine transform algorithm, control the laser device to form a laser-etched microstructure layer according to the embedded image data, and control the online imaging layer to acquire images and extract the digital watermark information 105 from the acquired images for verification.

[0026] The improved Discrete Cosine Transform (DCT) algorithm is a block-based DCT algorithm. It divides the image into multiple 8×8 pixel blocks and embeds the same 1-bit information "pairwise" only in the two low-to-mid-frequency coefficients (2,3) and (3,2) with the smallest JPEG quantization step size. The embedding order is determined by a global shuffle followed by writing, and the extraction is decided by a vote. The embedding rule is: for the DCT coefficient J of 8×8 block i,j, to write bits w∈{0,1}, J2,3←J2,3+α (2w 1), J3,2←J3,2+α (2w 1); Extraction rules are as follows =½[sgn (J2,3+J3,2)+1]. The basic embedding strength α is 2, linearly mapped to [0.8,2] by the block standard deviation σ, with the mapping formula α=0.8+1.2. σ / 25, where σ is the standard deviation of an 8×8 pixel block, σ≤5 for flat areas 0.8×, σ≥25 for texture areas 2×, ensuring that PSNR is always >40dB.

[0027] The watermark extraction algorithm uses ARM-Neon parallel 8×8 DCT operation, scanning the entire image in six milliseconds. It only accumulates the votes of the (2,3) and (3,2) coefficient symbols to obtain bits. The table lookup and reverse shuffling of 4096 bits takes 0.8ms, and the total extraction time is ≤16ms. The matching mechanism is as follows: the local 512-bit extracted sequence is encrypted with AES-CTR and sent to the cloud. It is XORed with the reference sequence to obtain the Hamming weight HD. HD≤18 is considered true, 19–30 is re-imaged, and >30 is directly rejected. The result is written back to MySQL as an HMAC-SHA256 digest for SPC to close the loop in one minute.

[0028] The digital watermark embedding and recognition system for wine packaging also includes a synchronization and calibration unit for achieving precise alignment between laser etching and printed graphics. The synchronization and calibration unit includes an encoder (1000 p / m) mounted on the drive side of the printing press, and a photoelectric sensor for detecting printing overprinting marks. The encoder's A phase is connected to the laser control card and resets using a "zero-speed pulse." After printing the first color group, the photoelectric eye captures a 1mm black overprinting mark (80mm spacing), and the rising edge serves as the watermark starting reference, triggering the laser to etch at 120µm intervals, ensuring the 8×8 DCT block and the printed grid are 1:1 without slippage. The first image taken by the machine is captured by a 300dpi linear CCD. Hough calculates the mark center. If the offset from the theoretical position is >0.05mm (1 pixel), the PLC fine-tunes the XY drive of the laser scanning mirror (±0.01mm step size) and corrects the encoder origin. The alignment error can be reduced to within 1 pixel after ≤3 closed-loop images. Subsequently, automatic sampling is performed every 500 images. If the cumulative offset is >0.02mm, recalibration is triggered. When the paper length expands by 0.9‰ due to long-term temperature drift, a second photoelectric eye is added 0.5m in front of the laser station for micro-slip compensation to ensure that the watermark and printed text maintain an alignment accuracy of ±0.05mm for a long time at a full speed of 120m / min.

[0029] This invention also provides a method for embedding and recognizing digital watermarks on wine packaging, applied to the aforementioned wine packaging digital watermark embedding and recognition system, comprising the following steps: Step S1 involves embedding the digital watermark information into a preset image of the wine packaging using an improved discrete cosine transform algorithm, generating watermarked image data. The embedding strength is adaptively adjusted based on the local texture complexity of the image blocks, using the formula α = 0.8 + k. σ / 25, where α is the embedding intensity, σ is the standard deviation of pixels within the block, and k is a constant.

[0030] Step S2: Based on the watermarked image data, control the laser equipment to etch a laser-etched microstructure layer corresponding to the changes in DCT coefficients on a designated layer of the printed packaging material. The laser etching parameters include: wavelength 355nm, average power 3.0W±0.05W, pulse frequency 50-70kHz, scanning speed synchronized with the production line speed of 120m / min, and each 8×8 pixel DCT block mapped to a 120µm×120µm physical etching cell.

[0031] Step S3: Use an online imaging device to capture images of the areas containing the laser-etched microstructure layer on the finished product packaging.

[0032] Step S4: Extract digital watermark information from the acquired image and match it with the original embedded information to verify its authenticity, outputting the authenticity determination result. The extraction algorithm calculates the DCT coefficients of each block in parallel, accumulates and votes on the symbols of two specified low-frequency coefficients to obtain the watermark bits, and determines authenticity by calculating the Hamming distance between the extracted sequence and the reference sequence, with a authenticity threshold ≤ 18.

[0033] A 64×64 block DCT embedding model was created using MATLAB, and anti-JPEG-70 compression tests were performed on 100 images each from three image libraries: white card, PET, and corrugated cardboard. The embedding strength α was calculated using the formula α = 0.8 + 1.2. σ / 25 is adaptively adjusted, embedding watermark information only in pairs at coefficients (2,3) and (3,2). Test results show that the bit error rate is <5%, meeting the design requirements, and the model is finalized.

[0034] Example 1: System Structure and Process Integration The digital watermarking system for wine packaging in this embodiment includes, from bottom to top: a substrate layer 101, such as 230gsm white cardboard; a base ink layer 102, printed using CMYK flexographic printing with a thickness of approximately 2μm; a laser-etched microstructure layer 103; a surface protective varnish layer 104; and a UV varnish with a thickness of approximately 1.5μm and a refractive index of 1.51. A linear CCD camera (300dpi) and a 30° side-lit LED unit for the online imaging layer are installed downstream of the laser etching station.

[0035] The system is integrated into the paper output section of a four-color flexographic printing press. An incremental encoder 301 (1000 pulses / meter) is installed on the drive side of the printing press. After the first color group is printed, the photoelectric sensor captures a preset 1mm black overprint mark. The rising edge signal of this mark serves as a trigger reference and is sent to the laser controller. The encoder's pulse signal is used to synchronize the etching cycle of the laser scanning mirror, ensuring that each encoder pulse corresponds to a 120μm physical etching step, corresponding to a "cell" mapped to an 8×8 pixel DCT block.

[0036] On the first print, the online imaging layer captures a pattern containing overprinting marks, and the actual center position of the marks is calculated using Hough transform. If the deviation from the theoretical position exceeds 0.05mm, the PLC control system fine-tunes the X / Y drive offset of the laser scanning mirror, completing the initial calibration within three prints. During production, an automatic inspection is performed every 500 prints. If the cumulative deviation exceeds 0.02mm, dynamic compensation is triggered to ensure long-term alignment accuracy.

[0037] Example 2: Watermark Embedding and Laser Etching The watermark embedding and physical implementation process is as follows: Watermark Generation and Embedding: The information to be embedded is first AES encrypted and globally shuffled, such as a 512-bit product serial number. For a pre-designed layout of the wine packaging, with a resolution of, for example, 720×1280, an improved DCT embedding algorithm is run in MATLAB or an equivalent environment. The algorithm divides the image into non-overlapping 8×8 pixel blocks. A DCT transformation is performed on each block. The standard deviation σ of the block is calculated according to the formula σ=std(8×8 block pixel values). The embedding strength α of the block is calculated according to the formula α=0.8+1.2·σ / 25, taking 0.8 when σ≤5 and 2.0 when σ≥25. Then, in the DCT coefficient matrix of the block, for the two coefficient positions (2,3) and (3,2) with a quantization step size Q=6, the same watermark bit w is embedded according to the rules J'2,3=J2,3+α·(2w-1) and J'3,2=J3,2+α·(2w-1), such as 0 or 1. After embedding, an inverse DCT transformation is performed to generate a visually normal watermarked digital image.

[0038] Laser etching physical implementation: The processing unit modifies the DCT coefficients of the watermarked image. For each 8×8 block, the coefficients (2,3) and (3,2) need to be increased by +α or decreased by -α, generating a laser etching control command queue. After receiving the command, the laser, with a wavelength of 355nm, an average power of 3.0W, and a pulse frequency of 60kHz, etches on the surface of the varnish layer corresponding to each "cell" (120μm×120μm).

[0039] If the watermark information of this block requires the corresponding DCT coefficient to increase by +α, representing a state of bit '1' or '0', then a 2×2 pit array is etched in the cell, with each pit having a depth of about 0.25μm and a diameter of about 4μm.

[0040] If the coefficient is required to decrease by -α to represent another state, then this cell is skipped.

[0041] For the identification of paired coefficients (2,3) and (3,2) sharing the same cell, they can be distinguished by rotating the entire pit array by 90°. In practical applications, a fixed mapping relationship can be chosen. After etching is completed, the UV lamp instantly cures the protective varnish layer.

[0042] Example 3: Online Identification and Verification Image Acquisition and Preprocessing: When the packaging with laser-etched microstructures passes through the in-line imaging layer, the encoder pulse triggers the side-lit LEDs for a 20μs strobe exposure, while the linear CCD simultaneously acquires the Cb channel grayscale image. The acquired image is first subjected to rapid perspective correction using Hough transform to detect paper edges or crease lines, and then a 31×31 pixel "white cap" filter is used to eliminate uneven illumination, improving contrast by approximately 18%. Processing a single frame takes approximately 4ms.

[0043] Watermark Extraction and Matching: The preprocessed image is fed into the processing unit, which uses the ARM-Neon instruction set for parallel acceleration. The algorithm divides the image into 8×8 pixel blocks according to a known 120μm grid, corresponding to the embedding process, and performs DCT transformation on each block. Only the DCT coefficient values ​​at positions (2,3) and (3,2) in each block are read and summed: S = J'2,3 + J'3,2. The sign of S (sgn(S)) is used to vote on whether the bit to be embedded in that block is 0 or 1. For example, =0.5*[sgn(S)+1]. After extracting bits from all blocks sequentially, a lookup table and reverse shuffling operation are performed to recover the original 512-bit encrypted sequence. The entire extraction process can be completed within 16ms.

[0044] Authenticity Verification and Feedback: The extracted encrypted sequence is sent to the cloud server via a TLS encrypted network. The cloud server decrypts the sequence using the same key and then performs a bitwise XOR operation with the original reference sequence stored in the database to calculate the Hamming distance (HD). The judgment rules are as follows: HD≤18, the item is deemed genuine, the system records the result and may display a green light; 19≤HD≤30, the item is deemed suspicious, the system may issue an alarm and request a second photo inspection; HD>30, the item is deemed counterfeit, and the system sends a signal to the rejection agency to remove it.

[0045] Closed-loop control: The system monitors the bit error rate (BER) in real time. For every 500 images detected, the average BER of the most recent 100 images is calculated. If the average BER exceeds a 3% threshold, the control logic automatically fine-tunes the laser's duty cycle, equivalent to power adjustment in steps of 0.05W up to a maximum of 3.6W, or adjusts the current of the side-light LEDs by +5%. It may also ignore the most vulnerable columns at the image edges, thus stabilizing the BER below 3% and ensuring a decoding success rate of over 99.7%.

[0046] Example 4: Application Effect Verification On a test platform simulating a production line environment, with a linear velocity of 2 m / s, inkjet printing was used to simulate the reflectivity differences caused by laser etching on a microporous membrane, thus constructing a verification system. The system ran continuously for 4 hours, testing 5000 tags. Results showed an average decoding success rate of 97.0%, an average error rate of 2.1%, and no system malfunctions. Consumers uploaded photos of the packaging via a WeChat mini-program, and the cloud returned verification results on average within 3 seconds. This preliminary verification demonstrates the superiority of the proposed solution in terms of recognition success rate, response speed, and system stability.

[0047] A watermark verification system based on an inkjet microporous membrane was constructed to simulate the microscopic scattering differences generated by laser etching. The core of the system is an Epson L805 pigment inkjet printer with a minimum droplet volume of approximately 1.5 pL, corresponding to an ink dot diameter of approximately 14 µm. A 2×2 ink dot array (approximately 28 µm × 28 µm) is printed on the surface of a PET microporous membrane with an 800 nm pore size and a thickness of 10 µm as "high scattering units," while the blank area serves as a "low scattering unit," creating a macroscopic reflectivity difference of approximately 0.3%. Tags are affixed to the outer edge of a 30 cm diameter turntable with a stable linear velocity of 2 m / s, equivalent to a 120 m / min production line. A 30° side-lit, 240 fps USB camera continuously captures images with 1 ms exposure. MATLAB performs real-time grayscale reading, 8×8 DCT, (2,3)(3,2) coefficient symbol voting, and shuffling. The average time from image capture to output for a single tag is 12 ms. The system ran continuously for 4 hours, testing 5000 tags with a decoding success rate of 97.0% and an average bit error rate (BER) of 2.1%. There was zero frame loss in the camera, and the equipment functioned without fault. A random sample of 200 tags was visually inspected by 10 people, and no pattern anomalies were observed that were perceptible to the naked eye. Throughout the entire experimental period, no ink cartridges were replaced or the printheads were cleaned. The experiment was repeated three times during the day, with the success rate fluctuating by only ±0.3%, verifying the high recognition rate, low error rate, and long-term stability of the solution.

[0048] 1. Create a 64×64 block DCT embedding model using MATLAB, and perform anti-JPEG-70 compression tests on 100 images each from three types of image libraries: white card, PET, and corrugated cardboard. The final version is considered to have a bit error rate of <5%. 2. Output 175 lpi flexible plates using an A3-size laser imagesetter, arrange 200 plates with different loads horizontally, and then print them on a laboratory screen printing machine (90 m / min). After printing, take a handheld 4-megapixel mobile phone photo, and the cloud decoding success rate is ≥95%. 3. The dual-head laser unit is hoisted to the last paper output section of the four-color press, and a 10 mm watermark strip is written synchronously at 120 m / min. Visual inspection is performed every 500 sheets, and the BER>3% is automatically increased by 0.1 level energy. A 20 mm coding area is left, and after die-cutting, social compression inspection is performed, with 93% of the paper being released. 4. The WeChat mini program launched the "Scan Code to Verify Authenticity" entry, which allows users to upload photos and receive verification results from the cloud, with an average processing time of less than 3 seconds. In the first month, 500 verification logs were collected to identify false rejection scenarios and fine-tune the threshold. 5. Retrain the embedding strength every quarter to gradually reduce the bit error rate to below 3%; simultaneously apply for group standards and complete the feasibility verification of low-cost DCT watermarking in the wine packaging industry. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital watermark embedding and recognition system for wine packaging, characterized in that, The digital watermark embedding and recognition system for wine packaging includes, from top to bottom: Substrate layer; A base ink layer is coated on the substrate layer to carry printed graphic information; A laser-etched microstructure layer is formed on the base ink layer, containing a microscopic physical structure corresponding to the embedded digital watermark information; A surface protective varnish layer is applied over the laser-etched microstructure layer to protect the microscopic physical structure. An online imaging layer is used to acquire images of the packaging area containing the laser-etched microstructure layer; The system further includes a processing unit, which is used to embed the digital watermark information to be embedded into the packaging image through a discrete cosine transform algorithm, control the laser device to form the laser-etched microstructure layer on the surface protective varnish layer or a designated layer according to the embedded image data, and control the online imaging layer to acquire images and extract the digital watermark information from the acquired images for verification.

2. The digital watermark embedding and recognition system for wine packaging according to claim 1, characterized in that, The laser-etched microstructure layer is formed by laser etching to a depth of 0.4-0.6µm, and physically writes 768 bits of information, including a 512-bit digital payload. The laser-etched microstructure layer is etched using a diode-pumped all-solid-state Nd:YVO4 laser, and the laser is a TEM. 00 The fundamental mode has a wavelength of 355nm, an average power of 2.5-3.5W, a pulse frequency of 50-70kHz, a single pulse energy of 50µJ, a pulse width of 15ns, a bidirectional grid scanning path, a line spacing of 6µm, and a scanning speed of 120m / min, which is synchronized with the printing production line; the microscopic physical structure is an array of pits formed in the region corresponding to each image block, with a pit depth of 0.2-0.3µm and a diameter of 3-5µm; In the laser-etched microstructure layer, each 8×8 pixel DCT block corresponds to a 120µm×120µm laser "cell". When the DCT coefficient needs to be +α, a 2×2 array of pits is etched in the cell, with a pit depth of 0.25µm and a diameter of 4µm. When –α is needed, it is skipped. Paired coefficients (2,3) and (3,2) share the same cell, and the pit array is rotated 90° to distinguish them.

3. The digital watermark embedding and recognition system for wine packaging according to claim 2, characterized in that, The discrete cosine transform algorithm is a block DCT algorithm, which divides the image into multiple 8×8 pixel blocks. In each block, the two low-frequency coefficients with the smallest quantization step size in the DCT domain are selected for pairwise embedding of watermark bits. The embedding strength is adaptively adjusted according to the standard deviation of the pixels in the block. The digital watermark embedding and recognition system for wine packaging uses an improved DCT algorithm to embed watermarks. It embeds the same 1-bit information in pairs only in the low-to-medium frequency coefficients (2,3) and (3,2) with the smallest JPEG quantization step size. The embedding amount α is linearly scaled with the local standard deviation σ of the image. The embedding order is written after global shuffling. The extraction is performed by voting. The embedding rule of the block-based DCT algorithm is as follows: for the DCT coefficients J of 8×8 blocks i and j, when writing bits w∈{0,1}, J2,3←J2,3+α (2w 1), J3,2←J3,2+α (2w 1); Extraction rules are as follows =½[sgn (J2,3+J3,2)+1]; The base value of the embedding quantity α is 2, and it is linearly mapped to [0.8, 2] by the block standard deviation σ, with the mapping formula α = 0.8 + 1.

2. σ / 25, where σ is the standard deviation of an 8×8 pixel block.

4. The digital watermark embedding and recognition system for wine packaging according to claim 3, characterized in that, The watermark extraction algorithm executed by the processing unit includes: performing block DCT transformation on the image acquired by the online imaging layer, accumulating the symbols of the two specified low-frequency coefficients and performing a voting decision to recover the watermark bit sequence, comparing the extracted sequence with a reference sequence in the cloud or locally, and determining the authenticity based on the Hamming distance; The image preprocessing of the online shooting layer includes: encoder triggering side light for 20µs exposure to obtain the Cb channel, perspective correction through Hough edge creasing lines, and smoothing out the vignetting with a 31×31 white cap. The watermark extraction algorithm in the online shooting layer is ARM-Neon parallel 8×8 DCT operation, accumulating coefficients (2,3)(3,2) symbol voting to obtain bits, looking up the table and reverse shuffling 4096 bits, with a total extraction time ≤16ms. The watermark matching mechanism is as follows: the local 512-bit extracted sequence is encrypted with AES-CTR and sent to the cloud. It is XORed with the reference sequence to obtain the Hamming weight HD. HD≤18 is considered true, 19–30 is a second image is taken, and >30 is directly rejected. The result is written back to MySQ as an HMAC-SHA256 digest.

5. The digital watermark embedding and identification system for wine packaging according to claim 4, characterized in that, The digital watermark embedding and recognition system for wine packaging also includes a synchronization and calibration unit for achieving precise alignment between laser etching and printed graphics. The synchronization and calibration unit includes an encoder installed on the drive side of the printing press and a photoelectric sensor for detecting printing overprint marks. The signal from the encoder is used to trigger the start and cycle of laser etching, and the mark position deviation detected by the photoelectric sensor is used to provide real-time feedback to adjust the coordinate origin of the laser scan.

6. The digital watermark embedding and identification system for wine packaging according to claim 5, characterized in that, The digital watermark embedding and recognition system for wine packaging maintains alignment accuracy through closed-loop control logic. The closed-loop control logic includes: periodically calculating the offset between the actual etching position and the theoretical position by acquiring images through the online imaging layer; if the offset exceeds a threshold, automatically adjusting the driving parameters of the laser scanning mirror or the encoder origin compensation value.

7. The digital watermark embedding and recognition system for wine packaging according to claim 6, characterized in that, The refractive index of the surface protective varnish layer is 1.50-1.52, which forms a refractive index difference with the air in the pits of the laser-etched microstructure layer, resulting in a shadow contrast of more than 8 gray levels under side lighting.

8. The digital watermark embedding and identification system for wine packaging according to claim 7, characterized in that, The substrate layer is a single-material white cardboard or biodegradable polylactic acid film, the base ink layer is a CMYK four-color flexographic ink layer with a thickness of 1-3μm, the surface protective varnish layer has a thickness of 1-2μm and is a UV-curable varnish, and the wine packaging digital watermark embedded identification system as a whole constitutes a packaging structure that can be recycled from a single material.

9. A method for embedding and recognizing digital watermarks on wine packaging, characterized in that, Applied to the system according to any one of claims 1-8, the method comprises: S1. The digital watermark information is embedded into the preset image of the wine packaging through the improved discrete cosine transform algorithm to generate watermarked image data. S2. Based on the watermarked image data, control the laser device to etch a laser-etched microstructure layer corresponding to the change of DCT coefficient on a designated layer of the printed packaging material; S3. Acquire images of the area containing the laser-etched microstructure layer on the packaging of the finished product using an online imaging device; S4. Extract digital watermark information from the acquired image, match and verify it with the original embedded information, and output the authenticity judgment result.

10. The method for embedding and recognizing digital watermarks on wine packaging according to claim 9, characterized in that, In step S1, the embedding intensity is adaptively adjusted according to the local texture complexity of the image block. The adjustment formula is: α = 0.8 + k·σ / 25, where α is the embedding intensity, σ is the standard deviation of pixels in the block, and k is a constant. The laser etching parameters in step S2 include: wavelength 355nm, average power 3.0W±0.05W, pulse frequency 50-70kHz, scanning speed synchronized with the production line speed of 120m / min, and each 8×8 pixel DCT block mapped to a 120µm×120µm physical etching cell. In step S4, the algorithm extracts and calculates the DCT coefficients of each block in parallel. The symbols of two specified low-frequency coefficients are accumulated and voted to obtain the watermark bits. The Hamming distance between the extracted sequence and the reference sequence is calculated to determine the authenticity, and the authenticity threshold is ≤18.

Images