Classification algorithm using halftone features of counterfeit bills and CNN.

Area of Science:

• Forensic Science
• Computer Vision
• Digital Image Processing

Background:

• Home printers are increasingly used for counterfeiting due to technological advancements.
• Identifying the printer model is essential for law enforcement to track criminals and solve forgery cases.
• Existing methods may be destructive or require specialized equipment.

Purpose of the Study:

• To propose a non-destructive technique for classifying printer models based on halftone textures.
• To differentiate between inkjet and laser printers and classify specific models.
• To develop an accessible and efficient method for forensic document examination.

Main Methods:

• Analysis of halftone textures in scanned documents printed by various home printers (12 color laser jets, 2 inkjets).
• Application of Fast Fourier Transform (FFT) to analyze periodic halftone lattices.
• Utilizing a Convolutional Neural Network (ResNet) for printer classification.
• Cropping images into parts for detailed texture analysis.

Main Results:

• 100% accuracy in distinguishing between laser and inkjet printers.
• 98.44% accuracy in classifying printer makes.
• Achieved 100% accuracy in model classification when analyzing 50 cropped images.
• The technique is non-destructive and requires only a scanner.

Conclusions:

• The proposed halftone texture analysis method is highly effective for printer model identification.
• This technique offers a practical, accessible, and efficient solution for forensic document analysis and combating counterfeiting.
• The method's non-destructive nature and reliance on standard scanners enhance its applicability in real-world investigations.