Anti-counterfeit software: the complete guide

The important observation is that these fail in different directions. A cloned code passes serialisation and fails image comparison. A pixel-perfect artwork copy on the wrong press passes image comparison and fails print forensics. A brand running only one of the three has a predictable blind spot, and counterfeiters find blind spots for a living.

Every unit gets a unique, non-sequential code that is validated against a live database. It is the most widely deployed approach in Indian trade categories because it doubles as a loyalty rail — the same scan that verifies also pays the tradesman, which is what generates the scan volume that makes the data useful. Covered fully in QR code programs and one QR doing both jobs.

Its limit is specific and worth stating plainly: a code photographed from a genuine pack and reprinted onto a thousand fakes will verify on every one of them. The defence is scan-count monitoring, which detects the clone only after it is already circulating. See QR program fraud prevention for how cloned codes are detected in practice.

This is the approach most brands have not evaluated, and it is the only one that works on stock already in the market. You photograph genuine samples, the system learns what genuine looks like statistically, and then any field photograph of a suspect pack can be checked against that model. No code, no label, no printing change, no packing-line downtime.

The design principle that makes it work is counter-intuitive: genuine is a distribution, not an image. Offset and rotogravure printing legitimately vary between units — colour-to-colour registration drifts ±0.1–0.2mm, ink density moves ΔE 2–4, substrate shade differs between reels, print-to-die-cut drifts ±1–2mm. A system that compares a suspect against one golden image will reject genuine product constantly. A system that learns the tolerance band from a set of genuine samples asks the right question: is this sample inside the envelope? The method is unpacked in image-based product authentication.

Halftone screening, screen angles and dot structure are properties of the press and the plate, not of the artwork. Two files that look identical on screen produce measurably different printed surfaces on different presses. In principle this is the hardest layer to defeat, because it requires the counterfeiter to match manufacturing rather than design.

In practice it is the layer most often unavailable, and honest vendors say so. Most phone cameras cannot resolve the screen at working distance; at ordinary capture resolution the frequency profile is dominated by the artwork rather than the screening. It contributes real evidence with a clip-on macro lens or a flatbed scan and very little without one. Detail in print forensics.

1. Onboards the genuine product — several photographs of known-good units, from which the tolerance envelope is learned.
2. Captures the suspect in the field, with live quality gates for sharpness, glare and steadiness, because a blurred photograph produces a meaningless verdict.
3. Measures — geometry against the learned bands, colour against learned ranges, print structure where resolution allows.
4. Reasons — a vision model extracts and compares specific features and adjudicates against the measurements.
5. Returns a verdict with evidence, not just a score: which elements differ, and by how much.
6. Aggregates — every verdict is a geolocated data point, and the map of those points is what brand protection actually acts on.

That last step is the one brands undervalue. A single verdict settles one argument; a thousand verdicts show you which three districts and which two distributors are the problem.

A system that only says genuine or fake will call a badly-lit photograph a counterfeit, and a brand-protection team that gets burned once by that stops trusting the tool. Separating I cannot tell from this is wrong is the difference between a system people use and one they quietly abandon.

Image-based authentication has the lowest barrier to trying: no packaging change, no line downtime, and a per-SKU setup measured in photographs rather than production cycles. That is why it is usually the right first step for a brand that suspects a problem but cannot yet justify serialising a catalogue.

• "99% accurate." On what test set, against which counterfeiter tier, at what capture quality? Accuracy on clean studio images is not accuracy on a WhatsApp-forwarded photograph.
• "No reference samples needed." Something has to define genuine. If not your samples, then a generic model that has never seen your pack.
• "Works on any product." Curved surfaces, transparent packaging, flat-colour artwork and textiles are all genuinely harder, and honest vendors will name their weak categories.
• "Detects counterfeits instantly." A real multi-layer verification takes tens of seconds. Instant usually means one shallow check.
• "Ready on day one." Thresholds tuned on synthetic or borrowed data need recalibrating on your actual packs. Treat the first fifty real verifications as calibration.

The buying framework is in choosing anti-counterfeit software, and the category question — verification-first or loyalty-first — in anti-counterfeit platform versus loyalty platform.