Banknote Image ROI Classification for Counterfeit Authentication

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for authenticating banknotes lack efficiency and accuracy in distinguishing genuine from counterfeit notes, particularly when using wavelet-based decompositions that do not adequately consider the significance of different sub-areas in the image analysis.

Innovation Solution

The method involves dividing the banknote image into regions of interest (ROIs), transforming each ROI into vectors using wavelet transformations and statistical moments, concatenating these vectors into a single vector representing the entire image, forming a matrix, selecting features based on predefined criteria, and using multiple classification models for authentication, with a consensus-based final classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If wavelet-based decomposition is used for image analysis, then measurement precision is improved, but device complexity increases

Engineering Contradiction:
Improveauthentication accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The image is divided into multiple regions of interest (ROIs) based on pre-defined criteria, where each ROI contains specific security features. This segmentation allows the complex wavelet analysis to be applied selectively to relevant areas rather than the entire image, improving authentication accuracy while managing processing complexity through focused analysis of critical regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The method extracts only the significant features from wavelet coefficients using statistical moments (mean, variance, skewness, kurtosis) and selects relevant wavelet sub-bands. This extraction process filters out redundant information and retains only the most discriminative features for authentication, thereby improving measurement precision while reducing the computational burden of processing all wavelet coefficients.

Inventive Principle:
Principle #2Taking out (Extraction)

2Reliability

If multiple classification models are used, then reliability is improved, but device complexity increases

Engineering Contradiction:
Improveauthentication reliabilityVSAvoidmodel complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

Multiple classification models (e.g., SVM, Random Forest, Neural Networks) are trained independently on different feature subsets and then merged through ensemble voting or averaging. This combination leverages the strengths of different algorithms to improve authentication reliability, while the modular structure allows each model to be trained separately, managing complexity through distributed processing.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The method performs preliminary feature selection and feature extraction before applying multiple classification models. By pre-processing the data to identify the most relevant features and organizing them into meaningful sets, the system reduces the complexity burden on subsequent classification stages, allowing multiple models to operate more efficiently on already-processed feature vectors.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If feature selection based on statistical moments is applied, then measurement precision is improved, but loss of information increases

Engineering Contradiction:
Improvefeature discrimination accuracyVSAvoidimage detail loss
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The method extracts specific statistical moments (mean, variance, skewness, kurtosis) from wavelet coefficients to characterize feature distributions. These moments capture essential discriminative information about the security features while filtering out redundant details. The extraction is performed selectively on significant wavelet sub-bands, preserving critical authentication information while discarding less relevant data.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The transformation of wavelet coefficients into statistical moment parameters changes the representation from raw pixel/intensity data to abstract statistical descriptors. This parameter transformation concentrates the essential authentication information into compact statistical measures, improving measurement precision by emphasizing discriminative patterns while reducing data dimensionality and associated information loss.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4500495B1Method for authenticating a banknote having at least one printed area on the substrate thereof
Publication Date: 2025.11.05 KOENIG & BAUER AG
  • EP4500495B1 patent drawingFigure 1
  • EP4500495B1 patent drawingFigure 2~3
  • EP4500495B1 patent drawingFigure 4~6

AI summary

The invention relates to a method for authenticating a banknote (01) having at least one printed area (02) on the substrate (03) thereof, wherein an optoelectronic capture device (06) captures the at least one printed area (02) on the particular substrate (03) of each of a plurality of copies (a; b; c) of the same type of banknote (01) to be authenticated and provides a processor unit (08) with raw image data for visualizing the particular captured printed area (02) in the form of an image (04) on a display device (09) that is connected to the processor unit (08) for data purposes, wherein the processor unit (08) partitions the particular image (04) into a plurality of ROIs and selectively provides different features for authenticating the particular type of banknote (01) from the raw image data relating to each of these ROIs after performing a plurality of mathematical operations, wherein differently selected features each form a classification model (KM), wherein a banknote (01) to be currently authenticated is authenticated using a plurality of classification models (KMx, where x = 1, 2, 3, ...).