Diffusion-Based Authentication of Copy Detection Patterns: A Multimodal Framework with Printer Signature Conditioning

Bolutife Atoki Iuliia Tkachenko Bertrand Kerautret Carlos Crispim-Junior · 2026

Read Diffusion-Based Authentication of Copy Detection Patterns: A Multimodal Framework with Printer Signature Conditioning by Bolutife Atoki Iuliia Tkachenko Bertrand Kerautret Carlos Crispim-Junior (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-08998.txt.

Introduction

Bolutife Atoki and colleagues check copy codes against counterfeits with a diffusion model. Sharp printing and generative AI now defeat old checks that struggle to separate fine counterfeits from real prints. Their tool jointly uses the binary master, the printed pattern, and a printer ID record, turning checks into multi-class printer sorting with spatial and text cues.

Catalog record

CREATORBolutife Atoki Iuliia Tkachenko Bertrand Kerautret Carlos Crispim-Junior
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39486
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.08998

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Diffusion-Based Authentication of Copy Detection Patterns: A Multimodal Framework with Printer Signature Conditioning Bolutife Atoki Iuliia Tkachenko Bertrand Kerautret Carlos Crispim-Junior Abstract 1 1 1 This paper has been accepted at WACV 2026. Counterfeiting affects diverse industries, including pharmaceuticals, electronics, and food, posing serious health and economic risks. Printable unclonable codes, such as Copy Detection Patterns (CDPs), are widely used as an anti-counterfeiting measure and are applied to products and packaging. However, the increasing availability of high-resolution printing and scanning devices, along with advances in generative deep learning, undermines traditional authentication systems, which often fail to distinguish high-quality counterfeits from genuine prints. In this work, we propose a diffusion-based authentication framework that jointly leverages the original binary template, the printed CDP, and a representation of printer identity that captures relevant semantic information. Formulating authentication as multi-class printer classification over printer signatures lets our model capture fine-grained, device-specific features via spatial and textual conditioning. We extend ControlNet by repurposing the denoising process for class-conditioned noise prediction, enabling effective printer classification. On the Indigo 1 × 1 1\times 1 Base dataset, our method outperforms traditional similarity metrics and prior deep learning approaches. Results show the framework generalizes to counterfeit types unseen during training. 1 Introduction Counterfeiting impacts industries from pharmaceuticals to food, posing serious health, safety, and economic risks. The widespread availability of high-quality printing and scanning devices, coupled with advances in generative deep learning methods, has driven the increase in counterfeit products. To address this challenge, some anti-counterfeiting solutions employ printable unclonable codes applied to products or packaging. A simple and cost-effective form of these codes is the Copy Detection Pattern (CDP), a maximum-entropy image that is sensitive to copying and undergoes information loss when reprinted [ 19 ] . However, advances in generative methods have introduced new challenges. Modern computer vision approaches [ 11 , 24 ] enable the estimation of binary approximations of CDPs, facilitating near-perfect replicas with little information loss, and undermine the reliability of traditional authentication methods [ 28 , 8 ] as illustrated in Figure 1 . In response, recent approaches have examined authentication using the binary template alone, printed CDP alone, or a combination of both. Independent studies [ 9 , 13 , 14 , 1 ] show printers leave unique, repeatable signatures on printed outputs due to subtle hardware and mechanical variations. These printer-specific signatures capture the identity of the printer, enabling discrimination even between different units of the same model. Unlike prior work using printer information only for grouping or evaluation, we treat the printer as an identity-bearing entity central to authentication. We represent each printer’s identity using a human-readable textual description, enabling the model to learn associations between signature patterns and printer classes, thereby supporting broader generalisation. We introduce a unified authentication framework that integrates the binary template, printed CDP, and printer identity in a single decision-making process. We formulate authentication as a multi-class classification task, where the goal is to identify the source printer. Our framework operates in the latent space using a diffusion-based architecture, which enables the model to capture low-dimensional representations that retain fine-grained, printer-specific details. Compared to prior generative approaches, Generative Adversarial Networks (GANs) [ 11 ] produce sharp images but often suffer from mode collapse and reduced diversity [ 2 ] , while Variational Autoencoders (VAEs) [ 16 ] cover more modes but typically yield blurrier results [ 4 ] ; diffusion models achieve both high fidelity and broad distribution coverage [ 10 ] . A textual description of the printer and a spatial conditioning mechanism guide the model’s attention to regions likely to contain these signature patterns. To perform classification, we introduce a novel classification block within a semantically and spatially conditioned setting, allowing the model to extract the printer’s signature from both the binary template and the printed CDP. Authentication is achieved by comparing the predicted printer identity with the expected one associated with the authentic CDP. Discrepancies between the two indicate potential counterfeiting. Figure 1 : Examples of a binary template, authentic printed CDP, and counterfeit printed CDPs from traditional image processing and deep learning methods. Normalised cross-correlation reliably detects traditional counterfeits but is unreliable against deep learning-based ones. In summary, our contributions are four-fold. First, we incorporate printer-specific signatures into the CDP authentication process, treating the printer as an identity-bearing entity rather than auxiliary metadata. Second, we propose the first unified authentication framework that jointly considers the binary template, printed CDP, and printer identity within a single pipeline. Third, we enable cross-printer generalisation by formulating authentication as a multi-class classification task over printer identities. Finally, we extend the image-conditioned diffusion model ControlNet [ 31 ] beyond its generative purpose by modifying its architecture to support classification. Specifically, we introduce a classification module that leverages the model’s learned denoising capabilities to perform class discrimination. Our implementation and models are available at our public repository 2 2 2 Repository URL is available at https://gitlab.liris.cnrs.fr/anr-trustit/diffusion_based_cdp_authentication.git . The remainder of this paper is as follows. In Section 2 , we review existing work on CDP authentication. Section 3 describes the proposed methodology and authentication strategy. In Section 4 , we perform experiments and the evaluation of our method. Section 5 concludes the paper with a summary of our findings and perspectives for future research. 2 Previous Work In this section, we review existing methods for authenticating copy detection patterns (CDPs). We begin by briefly discussing common counterfeiting techniques used to attack authentication systems, as understanding these attacks is essential for evaluating the effectiveness of various authentication approaches. Then, we broadly categorize existing authentication methods into three main approaches based on the type of data they utilise: (i) template-based methods, (ii) printer type-based methods, and (iii) printed CDP-based authentication methods. Counterfeiters estimate binary templates using techniques from thresholding (e.g., Otsu) to deep learning. This process is illustrated in Figure 1 , which shows the progression from a genuine binary template to an authentic printed CDP, followed by template estimation attacks and their printing as counterfeits. Models like Bottleneck DNN, autoencoders, SRGAN, and U-Net produce high-resolution approximations closely mimicking authentic templates [ 25 , 29 , 28 , 8 ] . Given the variety and sophistication of these counterfeiting techniques, authentication methods have been developed to detect and prevent forgery. These methods can be broadly classified based on the type of data they utilise, as detailed below. Template-based CDP authentication. The simplest form of template-based authentication compares a test printed CDP (probe) to its corresponding binary template using similarity metrics such as Normalised Cross-Correlation (NCC) or Structural Similarity Index (SSIM). A similarity score is computed and evaluated against a threshold derived from authentic pairs. These methods once sufficed but now fail against high-quality template estimates [ 3 , 8 , 28 ] . To improve reliability, more advanced approaches have been proposed. The authors in [ 26 ] introduced a method that models the printing-and-digitisation (P&D) process as a transformation of local 3\times 3 binary neighbourhoods. For each printer, a codebook is constructed, mapping the probability of bit-flip for every one of the 512 possible neighbourhood configurations. At authentication, the flipped bits in the binarised probe are compared to the expected flipping probabilities, and authentication is performed via thresholding.

Read the complete text