Neural Network Spoof Detection Using Distance Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image classification techniques fail to distinguish between real three-dimensional objects and their two-dimensional spoofs, posing security risks in applications like login procedures, as they are not designed to differentiate between the actual object and its representation.
Innovation Solution
A computer-implemented method using a neural network on a mobile device that processes images by calculating a distance map and reflection pattern, comparing them to learned patterns to determine if the image contains a real or spoofed object, thereby enhancing security by reliably identifying the authenticity of objects in images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional image classification techniques are used, then the system is simple and easy to implement, but it cannot distinguish between real three-dimensional objects and two-dimensional spoofs, resulting in poor security reliability
Solution Approach 1:
The patent transitions from conventional two-dimensional image classification to three-dimensional spatial analysis by calculating distance maps that represent the depth information of pixels relative to the optical sensor. This dimensional enhancement enables the system to detect whether an object is truly three-dimensional or a two-dimensional spoof, thereby improving security reliability without requiring an entirely new system architecture.
Solution Approach 2:
The patent introduces distance maps and reflection patterns as intermediary representations that mediate between the raw image data and the final authentication decision. These intermediaries capture spatial and optical characteristics that conventional image classification misses, enabling reliable distinction between real objects and spoofs while maintaining system implementability through existing mobile device hardware.
2Measurement precision
If advanced neural network processing is applied to distinguish real objects from spoofs, then authentication accuracy improves, but computing resource requirements increase
Solution Approach 1:
The patent segments the authentication task into distinct computational steps: extracting distance maps for each pixel, calculating reflection patterns, and comparing these features against learned patterns. This segmentation allows the system to process images efficiently using mobile device capabilities while maintaining high authentication accuracy through targeted computational focus on the most discriminative features.
Solution Approach 2:
The patent transforms the image processing approach by changing the parameter space from conventional pixel intensity analysis to spatial distance and reflection pattern analysis. This parameter transformation enables the neural network to operate with reduced computational complexity by working with derived features that directly capture the essential differences between real objects and spoofs, rather than processing the entire image data volume.
3Reliability
If distance map and reflection pattern calculation is performed, then the ability to detect spoofs improves, but processing time increases
Solution Approach 1:
The patent performs preliminary calculations of distance maps and reflection patterns during the image processing stage, preparing these features in advance for the authentication decision. By computing these spatial and optical characteristics upfront rather than during the final authentication check, the system reduces the time required for the critical authentication decision while maintaining comprehensive spoof detection capability.
Data Source
AI summary
A method for distinguishing a real three-dimensional object from a two-dimensional spoof of the real three-dimensional object, the method comprising: obtaining, by an optical sensor of a mobile device, an image containing an object that is either the two-dimensional spoof or the real three-dimensional object; providing the image to a neural network; processing the image by the neural network by calculating: 1) a distance map representative of the distance of pixels to the optical sensor, the pixels constituting at least a portion of the object within the image, or 2) a reflection pattern representative of light reflection associated with pixels constituting at least a portion of the object within the image; comparing the distance map or the reflection pattern with a learned distance map or a learned reflection pattern; and obtaining as a final output a determination the image contains either the two-dimensional spoof or the real three-dimensional object.