Fraudulent Image Detector for Single-Frame Identity Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting fraudulent portrait photo swaps in identity documents are inefficient, particularly when holograms or overlays are near pale or colorless, and often require multiple frames for analysis.
Innovation Solution
A fraudulent image detector using convolutional neural networks to analyze a single image, segmenting foreground and backdrop regions, and computing probabilities for security patterns, enabling detection of fraud by analyzing pixel-level and regional inconsistencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If chromaticity and hue-based heuristics are used for fraud detection, then detection capability is improved, but multiple frames are required and performance deteriorates when holograms are pale or colorless
Solution Approach 1:
The image is segmented into foreground and backdrop regions using a second segmenter, allowing independent analysis of each region. This segmentation enables the system to focus computational resources on relevant areas and make detection decisions based on regional inconsistencies rather than requiring multiple frames for temporal analysis
Solution Approach 2:
The invention transitions from temporal analysis (multiple frames) to spatial analysis (single image with regional segmentation). By computing probability maps across different spatial regions and comparing their statistical properties, the system achieves fraud detection in a single frame by exploiting spatial dimensionality instead of temporal dimensionality
2Reliability
If chromaticity and hue-based heuristics are used for fraud detection, then detection capability is improved, but detection accuracy deteriorates when holograms are pale or colorless
Solution Approach 1:
The system changes the analysis parameters from chromaticity and hue (color-based) to spatial distribution and statistical properties of probability data. By analyzing the spatial arrangement and statistical characteristics of security pattern probabilities rather than color properties, the system becomes insensitive to the color intensity or presence of holograms
Solution Approach 2:
Different regions (foreground and backdrop) are analyzed with different probability computations and statistical measures. The system applies region-specific probability maps and compares their local statistical properties, allowing detection based on qualitative differences in pattern distribution rather than uniform color characteristics
3Reliability
If pixel-wise chromaticity statistics are analyzed across multiple frames, then fraud detection is possible, but processing complexity and time consumption increase
Solution Approach 1:
The invention extracts only the essential features needed for fraud detection from the image data - specifically, the spatial distribution and statistical properties of security pattern probability maps in foreground and backdrop regions. By extracting and comparing these key statistical features rather than analyzing all pixel data across multiple frames, the system reduces processing complexity while maintaining detection reliability
Data Source
Figure 1~3
Figure 4
AI summary
The invention concerns a fraudulent image detector (4) comprising: a first segmenter (6) configured to compute first probability data from an image showing personal identifiable information of an individual, wherein the first probability data indicates, for each pixel of the image, a probability that the pixel shows a security pattern; a second segmenter (8) configured to compute second probability data from the image, wherein the second probability data indicates, for each pixel of the image, a probability that the pixel is part of a foreground region showing the personal identifiable information or part of a backdrop region showing no personal identifiable information; a classifier (12) configured to compute score data from the first probability data and the second probability data.