Privacy-Preserving Image Processing for Retail Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional point of sale systems are susceptible to fraud and errors due to operator negligence or malpractice, such as 'pass-throughs' and 'sweethearting,' and surveillance systems capture personal information that is vulnerable to breaches.
Innovation Solution
Transform image data to obscure personal information while preserving activity recognition by applying selective or whole-scene transforms to remove identifiable features, enabling secure monitoring without revealing sensitive details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If surveillance systems capture images in retail environments to detect fraudulent activities, then the ability to detect fraud is improved, but personal information becomes vulnerable to breaches
Solution Approach 1:
The system extracts and removes personally identifiable information (such as faces, license plates, or other identifying features) from surveillance images while retaining the contextual information needed for fraud detection. This is achieved through selective masking or blurring of specific regions in the image that contain personal information, allowing the system to maintain reliability in detecting fraudulent activities without capturing vulnerable personal data.
Solution Approach 2:
Different regions of the image are treated differently: regions containing personal information are obscured or transformed, while regions containing activity information are preserved. This local differentiation allows the system to simultaneously protect personal information and maintain fraud detection capability by applying quality transformations selectively to different parts of the image.
2Object-affected harmful factors
If image transforms are applied to remove personal information, then personal information security is improved, but the ability to recognize activity may be degraded
Solution Approach 1:
The system introduces an intermediary processing stage that transforms personal information into an intermediate representation that preserves activity patterns but removes identifying features. This intermediary form allows activity recognition to continue while personal information is protected, serving as a bridge between security requirements and detection needs.
Solution Approach 2:
The system changes specific parameters of the image (such as blurring radius, masking regions, or transformation intensity) to achieve the optimal balance between protecting personal information and preserving activity recognition. By carefully controlling the degree and type of transformation applied, the system minimizes information loss while maximizing security.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A processing resource receives original image data by a surveillance system. The original image data captures at least private information and occurrence of activity in a monitored region. The processing resource applies one or more transforms to the original image data to produce transformed image data. Application of the one or more transforms sufficiently distorts portions of the original image data to remove the private information. The transformed image data includes the distorted portions to prevent access to the private information. However, the distorted portions of the video include sufficient image detail to discern occurrence of the activity in the retail environment.