Identification of counterfeit collectible cards

A method for use in determining if a collectible card is counterfeit, the method comprising the steps of:

• maintaining a database of reference card images for known collectible card specifications;
• receiving, from a client device, a digital image of a subject collectible card;
• selecting a plurality of predetermined regions of interest from the digital image of the subject collectible card;
• identifying a matching reference card image from the database based on the collectible card specification;
• aligning the selected regions of interest of the subject card image with corresponding regions of interest of the reference card image using Scale-Invariant Feature Transform (SIFT);
• amplifying pixel data of the aligned regions of interest;
• comparing, via a neural network architecture that includes two identical sub-networks, the amplified regions of interest of the subject collectible card with the amplified regions of interest of the reference card by generating feature vectors and computing a distance metric between them;
• and
• determining whether the subject collectible card is authentic or counterfeit by classifying the card based on whether the computed distance metric satisfies a predetermined threshold.

A computer-implemented counterfeit detection system for collectible trading cards, the system comprising:

• one or more processors and at least one non-transitory computer-readable medium storing executable instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
• maintaining a database of authenticated reference card images indexed according to collectible card specifications;
• receiving, from a remote client device, a digital image of a subject collectible card;
• determining a collectible card specification of the subject collectible card and retrieving a corresponding authenticated reference card image from the database;
• selecting a plurality of predetermined regions of interest associated with the determined card specification from the subject collectible card image;
• aligning the selected regions of interest of the subject collectible card image with corresponding regions of interest of the authenticated reference card image using Scale-Invariant Feature Transform (SIFT);
• amplifying pixel data within each aligned regions of interest by repeating pixel values to produce amplified ROI data;
• processing the amplified regions of interest data through a neural network architecture comprising two identical sub-networks to generate a first feature vector from the subject collectible card regions of interest and a second feature vector from the authenticated reference card regions of interest;
• computing a distance metric between the first and second feature vectors;
• and
• outputting an authenticity determination that classifies the subject collectible card as authentic or counterfeit based on whether the computed distance metric satisfies a predetermined threshold.