📸 Photo Privacy Analysis

Overview

The Photo Privacy Analysis feature demonstrates the extensive personal information that can be extracted from your photos. This educational tool helps you understand why photo library access is a significant privacy consideration.

1. Basic Photo Analysis

Service: PhotoAnalysisService

What is Analysed

• Location Data — GPS coordinates embedded in photos reveal where you've been
• Temporal Patterns — When you take photos shows your daily routines
• Face Detection — Number of faces and groupings indicate social connections
• Device Information — Camera model, settings, and software versions
• Photo Types — Screenshots, selfies, panoramas, Live Photos distribution

Privacy Insights Generated

• Home and work location estimation
• Travel history and frequent destinations
• Social circle size estimation
• Activity patterns (morning person vs night owl)
• Device upgrade history

2. Extended Photo Profiling

Service: ExtendedProfilingService with 12 specialised analysis actors

Deep analysis that demonstrates the full extent of what can be learned from your photo library.

Analysis Categories

Device-Only Filtering

The analysis focuses on photos taken on your device, filtering out:

• Downloaded images and memes
• Screenshots (analysed separately)
• Photos received via messaging apps
• Stock photos and web downloads

Smart Sampling

For large photo libraries, the system uses intelligent sampling:

• Maximum 3,000 photos analysed per session
• Even time distribution — samples across your entire photo history
• Priority weighting — favours photos with rich metadata
• Batch processing — processes in groups of 50 for efficiency

3. What Each Category Reveals

🏃 Activities & Hobbies

Uses Vision framework to detect:

• Sports (running, cycling, swimming, gym equipment)
• Outdoor activities (hiking, camping, skiing)
• Creative hobbies (painting, music instruments, crafts)
• Gaming and entertainment preferences

👥 Social Patterns

• Top 5 People — Shows thumbnails of the people you photograph most (from iOS People album)
• Group photo frequency and sizes
• Recurring faces (close relationships)
• Social event attendance patterns
• Family vs friends vs colleagues estimation

🎉 Life Events

Detects significant moments:

• Weddings and engagements
• Graduations and ceremonies
• Birthdays and celebrations
• New baby / pregnancy
• Moving house / new home

🚗 Vehicles

• Car make and model detection
• Vehicle ownership patterns
• Transportation preferences
• Parking locations (home, work)

🍕 Food & Dining

• Restaurant visit frequency
• Cuisine preferences
• Home cooking vs eating out ratio
• Dietary patterns (vegetarian indicators, etc.)

📄 Sensitive Documents

Uses OCR to detect photos of:

• ID cards and passports
• Credit cards and bank statements
• Medical documents
• Contracts and legal documents
• Receipts with personal information

🏠 Home Content

• Room types and layout
• Furniture and appliances
• Home value estimation indicators
• Smart home devices visible

💼 Work Patterns

• Office environment detection
• Work schedule patterns
• Business travel frequency
• Industry indicators

4. On-Device AI Model

PrivacyInsights includes a bundled machine learning model that runs entirely on your iPhone. No photos or data are ever sent to a server.

🧠 MobileNetV3 Large

The app ships with MobileNetV3 Large, a CoreML image classification model trained on the ImageNet dataset with 1,000 object categories. This model powers the object and scene recognition used across all 12 analysis categories.

How It Works

• Image input — Each photo is resized to 224×224 and fed to the model
• Classification — MobileNetV3 identifies objects, scenes, animals, vehicles, food, and more from 1,000 categories
• Confidence filtering — Only results above a configurable confidence threshold are kept
• Category mapping — Classifications are mapped to the 12 privacy analysis categories (activities, vehicles, food, etc.)
• Fallback — If the model is unavailable, Apple's built-in Vision classification is used instead

5. Technical Implementation

Core Technologies

• Photos framework — Photo library access and metadata
• Vision framework — Image classification and object detection
• VNRecognizeTextRequest — OCR for document detection
• VNClassifyImageRequest — Scene and object classification
• CoreML — On-device machine learning

Actor-Based Architecture

Each analysis category runs as a separate Swift actor for thread safety:

• ActivityDetectionService — Activities and hobbies
• SocialPatternService — Social analysis
• LifeEventService — Life event detection
• VehicleDetectionService — Vehicle identification
• FoodDetectionService — Food and dining
• SensitiveDocumentService — Document OCR
• HomeAnalysisService — Home content
• WorkAnalysisService — Work patterns

Performance Optimisations

• Batch processing — 50 photos per batch with Task.yield()
• Thumbnail analysis — Uses 300×300 thumbnails, not full resolution
• Result caching — Stores results to avoid re-analysis
• Cancellation support — Can stop analysis at any time
• Progress tracking — Real-time progress updates

6. Privacy Implications

This analysis demonstrates why photo library access is one of the most sensitive permissions on iOS:

What Apps Could Learn

Data Type	Privacy Risk
Home location	Physical security, stalking risk
Work location	Employment information, schedule
Social connections	Relationship mapping, social engineering
Financial documents	Identity theft, fraud
Health information	Insurance discrimination, blackmail
Travel patterns	Burglary timing, location tracking
Vehicle information	Theft targeting, tracking
Children's photos	Family information, school locations

Recommendations

• Use "Selected Photos" — iOS lets you grant access to specific photos only
• Review permissions regularly — Check which apps have photo access
• Delete sensitive documents — Don't keep photos of IDs and cards
• Disable location in photos — Turn off geotagging for sensitive photos
• Be selective — Only grant photo access to apps that truly need it