Embedded Image Authentication for Printed Copy Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anti-copying technologies are ineffective due to the ease with which complex watermark patterns can be restored and counterfeited by high-precision printing devices and image-processing software, leading to reduced accuracy in identifying counterfeit products.
Innovation Solution
A method and apparatus for detecting copy products using an image information extraction model trained on sample printed and copy products to extract target image information, which is then compared with standard embedding information to determine if a product is a copy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a complex watermark pattern is added to a product before printing, then the product can carry identification information for counterfeit detection, but the watermark pattern becomes easier to restore and counterfeit with high-precision printing devices and image-processing software, reducing detection accuracy
Solution Approach 1:
The patent extracts only the necessary embedding information from the complex watermark pattern through an image information extraction model. This model processes the captured image to isolate and extract the embedded identification data, separating it from the rest of the product image. By extracting only the essential information rather than relying on the complete complex pattern, the system maintains detection accuracy while reducing vulnerability to counterfeiting attempts.
Solution Approach 2:
The patent introduces an image information extraction model as an intermediary between the captured image and the detection process. This model acts as a mediator that processes the raw image data, extracts the embedded information, and prepares it for comparison with standard data. The intermediary model enhances the robustness of detection by filtering out noise and counterfeiting artifacts while preserving the essential embedding information.
2Reliability
If a complex watermark pattern is used for product identification, then the product can be traced and authenticated, but the complexity of the pattern makes it more susceptible to restoration and counterfeiting by advanced printing and software tools
Solution Approach 1:
The system extracts only the essential authentication information from the complex watermark pattern through the image information extraction model. By taking out only the necessary embedding data rather than processing the entire complex pattern, the system maintains authentication reliability while reducing the impact of pattern complexity and its associated vulnerabilities to counterfeiting.
Solution Approach 2:
The patent replaces the mechanical/visual inspection of complex watermark patterns with an automated image information extraction model. This substitution transforms the authentication process from manual pattern recognition to automated information extraction, improving reliability by consistently identifying the embedded information regardless of pattern complexity or counterfeiting attempts.
3Ease of operation
If traditional watermark-based detection methods are used, then the detection process is simple and straightforward, but the accuracy of identifying counterfeit products decreases due to the ease of restoring watermark patterns
Solution Approach 1:
The patent introduces an image information extraction model as an intermediary layer between the simple capture process and the detection decision. This model maintains the simplicity of the overall process (capture-extract-compare) while significantly improving accuracy by intelligently extracting and processing the embedding information, filtering out counterfeiting artifacts, and preparing reliable data for comparison with standard embedding information.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A method for detecting a copy product includes that: an image of a to-be-detected product and standard embedding information are acquired (S101, S201, S304, S406); information extraction is performed on the image through an image information extraction model to obtain target image information in the image, where the image information extraction model is obtained by training according to an image of a sample printed product, sample embedding information of the sample printed product, and an image of a sample copy product, and the sample copy product is obtained by copying the sample printed product (S102, S202, S305, S407); and whether the to-be-detected product is the copy product is detected according to the target image information and the standard embedding information (S103, S306, S408).