Design Privacy with Analogia Graph

Yang Cai, Joseph Laws, Nathaniel BauernfeindAAAI 2010 · DOI

Abstract

Human vision is often guided by instinctual commonsense such as proportions and contours. In this paper, we explore how to use the proportion as the key knowledge for designing a privacy algorithm that detects human private parts in a 3D scan dataset. The Analogia Graph is introduced to study the proportion of structures. It is a graph-based representation of the proportion knowledge. The intrinsic human proportions are applied to reduce the search space by an order of magnitude. A feature shape template is constructed to match the model data points using Radial Basis Functions in a non-linear regression and the relative measurements of the height and area factors. The method is tested on 100 datasets from CAESAR database. Two surface rendering methods are studied for data privacy: blurring and transparency. It is found that test subjects normally prefer to have the most possible privacy in both rendering methods. However, the subjects adjusted their privacy measurement to a certain degree as they were informed the context of security.

Introduction

Modern security systems, such as millimeter-wave scanners used in airports, are essential for detecting non-metallic threats like plastic explosives or ceramic weapons. However, these scanners produce high-resolution 3D point clouds that reveal intimate anatomical details, leading to significant privacy concerns. Technical challenges in protecting this data include the "noisy" nature of 3D scans and the high diversity of human body shapes. Traditional computer vision methods often struggle with the speed required for busy security checkpoints or lack the robustness to handle non-standard poses. This paper introduces a fast, robust algorithm that leverages "Analogia Graphs"—mathematical representations of human proportions—to identify and obscure private regions in real-time.

The core innovation of this work is the Analogia Graph, a graph-based representation of intrinsic human proportions. By leveraging the artistic "commonsense" that an adult is roughly 6–8 heads tall, the algorithm uses the head as a universal ruler to navigate the 3D space.

The Analogia Graph uses human proportions to define search regions.

Instead of searching the entire 3D point cloud, the system uses these proportions to predict the location of "regions of interest." For example, based on anatomical norms, the chest is typically located 2–3 head-units from the top of the head. This targeted approach reduces the search space by an order of magnitude, allowing the system to bypass the need for exhaustive coordinate-based searches.

Methodology: Slicing and Template Matching

To further optimize computation, the 3D point cloud is segmented into 2D horizontal contours. This dimensionality reduction allows for much faster processing without losing the critical geometric features needed for identification.

3D body scans are segmented into 2D horizontal contours for efficient processing and privacy rendering.

The system then uses Radial Basis Functions (RBF) and non-linear regression to fit a mathematical template to these contours. By calculating invariant features like "Height Ratios" and "Area Ratios," the system remains robust regardless of the subject's absolute height or distance from the scanner. Once private areas are detected, they are obscured using either blurring (reducing polygon resolution) or transparency to protect the subject's dignity.

Key Results and Usability

The algorithm was validated using 100 datasets (50 male, 50 female) from the CAESAR database, achieving a success rate of over 90% in classifying body features. Notably, the system is 60 times faster than previous landmark-extraction records, making it viable for high-throughput security environments.

A usability study involving 20 subjects revealed that privacy preferences are highly contextual. While subjects initially preferred maximum levels of blurring, their willingness to accept less privacy increased significantly when informed of security risks, such as the presence of a hidden weapon. The average preferred blurring level dropped from ~4.4 to ~2.8 on a 5-point scale when the security context was introduced, highlighting a clear trade-off between privacy and safety.

Conclusion and Future Work

The Analogia Graph provides a powerful framework for integrating human-centric "commonsense" into automated privacy protection. While highly effective for standard standing poses, future work will focus on developing more robust methods to detect a wider range of body features and fine-tuning the balance between privacy and threat detection. The goal is to ensure that while intimate details are protected, the system remains capable of identifying small, concealed threats.

Summary updated

Abstract

Human vision is often guided by instinctual commonsense such as proportions and contours. In this paper, we explore how to use the proportion as the key knowledge for designing a privacy algorithm that detects human private parts in a 3D scan dataset. The Analogia Graph is introduced to study the proportion of structures. It is a graph-based representation of the proportion knowledge. The intrinsic human proportions are applied to reduce the search space by an order of magnitude. A feature shape template is constructed to match the model data points using Radial Basis Functions in a non-linear regression and the relative measurements of the height and area factors. The method is tested on 100 datasets from CAESAR database. Two surface rendering methods are studied for data privacy: blurring and transparency. It is found that test subjects normally prefer to have the most possible privacy in both rendering methods. However, the subjects adjusted their privacy measurement to a certain degree as they were informed the context of security.
BibTeX
@inproceedings{cai2010design,
  author = {Yang Cai and Joseph Laws and Nathaniel Bauernfeind},
  title = {{Design Privacy with Analogia Graph}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2010},
  doi = {10.1609/aaai.v24i2.18810},
}

Loading related content...

Figure