Autocorrelation-based Fiducial Markers for Traceability

Key idea:

• Fiducial markers are embedded invisibly within the autocorrelation of the image.

• Even small image crops can be rectified using these statistical, spatially invariant properties.

• Affine transformations are estimated from characteristic structures (e.g., peaks) in the autocorrelation.

• Translation is recovered independently using phase correlation.

Protecting French Wines from Parallel Resale

France, a global symbol of excellence in winemaking, is facing a growing challenge: controlling the distribution and pricing of its wine bottles. Many wines are sold to different clients (exporters, distributors, partners) under varying pricing conditions.

However, some actors exploit these price differences: 

• they purchase bottles at lower prices

• and resell them at higher prices in other markets

This grey market activity directly impacts producers, damages brand value, and disrupts carefully managed distribution strategies.

Today’s traceability systems rely mainly on visible identifiers such as:

• QR codes

• barcodes

But these solutions have a critical weakness: they can be easily damaged, removed, or falsified.

In practice, simply tearing or degrading these codes is enough to make a bottle untraceable, enabling unauthorized resale.

We introduce a new generation of marking based on invisible, embedded codes directly integrated into the label’s texture.

Unlike traditional identifiers:

• they cannot be easily removed or torn

• they remain detectable even after wear or deformation

• they are invisible to the naked eye, making them harder to target

To provide French wine producers with a discreet, robust, and secure traceability solution that:

• protects against grey market diversion

• preserves brand value and pricing integrity

• ensures product authenticity

A technology designed to meet the highest standards of the French wine industry. 

Geometric Challenge, Curved Surfaces and Texture Rectification

Wine bottles are inherently curved surfaces, which introduces significant geometric distortion when capturing images of their labels.
Unlike flat objects, textures wrapped around cylindrical or complex shapes undergo non-uniform deformations that cannot be corrected using simple global transformations.

This makes texture rectification particularly challenging:

• standard methods assume planar surfaces

• a single global homography is often insufficient

• local distortions vary across the surface

To address this, we rely on a local geometric modeling approach:

• The deformation is approximated locally by affine transformations (local affinities)

• These local observations are aggregated to estimate a global rectifying homography

• The method solves an inverse problem, reconstructing the deformation from local differential cues rather than explicit correspondences

This approach enables:

• robust rectification of textures on curved surfaces

• accurate decoding of embedded invisible markers

• resilience to partial observations and real-world imaging conditions