Biometric Portraits With Generative Art

ISEF Category: Technology Enhances the Arts

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your body leaves a data trail every second. Heart rate, skin sweat, and breathing all shift when your state changes. This project turns that signal trail into art. The real question is whether the same emotional state creates the same portrait twice.

What Is It?

This project links body signals to generative art. You measure heart rate, galvanic skin response, and breathing rate, then feed those features into an image model that creates a portrait. Think of it like a musical instrument, but instead of notes, your body gives the model control signals.

Heart rate from the MAX30102 sensor, skin response from simple electrodes and an ADS1115 converter, and breathing rate from a chest or microphone-based sensor all act like sliders. When the sliders change, the output image changes too. Your job is to test whether the system behaves in a stable way, especially when the same person tries to match the same emotional state on different days.

Why This Is a Good Topic

This is a strong science fair topic because you can test both engineering and human variability. You get a clear input-output system, which makes the project measurable. You also connect to a real problem, emotion-aware creative tools, while learning sensor calibration, signal processing, and similarity analysis. A student can start with a simple version, then add better controls and stronger analysis as the project grows.

Research Questions

• How does the same self-reported emotional state affect portrait similarity across different days?
• What is the effect of using heart rate alone versus using heart rate, skin response, and breathing together on portrait consistency?
• Does normalization of biometric signals improve the repeatability of generated portraits?
• To what extent do different prompt templates change the stability of the image output for the same biometric input?
• Which biometric feature, heart rate, skin response, or breathing rate, contributes most to variation in the final portrait?
• How does short-term stress before data collection affect similarity between portraits from the same emotional label?

Basic Materials

• MAX30102 heart-rate sensor module.
• ADS1115 analog-to-digital converter board.
• Homemade skin conductance electrodes.
• Breathing sensor such as a respiratory belt, stretch sensor, or microphone setup.
• Microcontroller or single-board computer such as Arduino or Raspberry Pi.
• Laptop with Python installed.
• Stable Diffusion or similar image-generation setup.
• Image comparison tool for visual similarity scoring.
• Notebook or spreadsheet for labeling sessions and emotions.

Advanced Materials

• Research-grade ECG or PPG sensor for heart-rate validation.
• Clinical or lab-grade galvanic skin response amplifier.
• Respiratory belt transducer with synchronized data output.
• GPU workstation for repeated image generation runs.
• Data acquisition hardware with timestamp synchronization.
• Standardized emotion induction materials approved by a supervisor or review board.
• Statistical software for mixed-effects modeling.
• Image similarity analysis package for embedding-based comparison.

Software & Tools

• Python: Processes sensor signals, controls the generation pipeline, and organizes trial data.
• ImageJ: Measures visual features such as color balance, brightness, and texture in portraits.
• OpenCV: Helps you inspect frames, align images, and extract image-level metrics.
• pandas: Stores session data in tables and makes comparison across days easier.
• Stable Diffusion: Generates the portrait images from your biometric input and prompts.

Experiment Steps

1. Define the emotional states you can reliably ask a participant to reproduce, such as calm, focused, or stressed.
2. Choose the biometric features you will trust as inputs, and decide how you will clean noisy sensor data.
3. Build a mapping from sensor values to image controls, then write down the exact rules so every session uses the same logic.
4. Plan a repeatability test that compares portraits made from the same emotional label on different days.
5. Select one image similarity metric and one human-rating method so you can compare algorithmic and visual consistency.
6. Set controls for lighting, prompt wording, and session timing so they do not mask the biometric effect.

Common Pitfalls

• Letting sensor contact change from session to session, which makes the biometric input reflect setup quality instead of emotion.
• Using an emotion label that the participant cannot reproduce reliably, which weakens the repeatability test.
• Changing the prompt wording between trials, which adds image variation that has nothing to do with the body data.
• Comparing portraits by eye only, which makes the results subjective and hard to defend.
• Ignoring signal lag between breathing, skin response, and heart rate, which can misalign the biometric features with the generated image.

What Makes This Competitive

A competitive version shows that you understand both the art system and the measurement problem. You would need tight controls, repeated trials, and a clear similarity metric, not just cool-looking outputs. Strong projects also test whether one biometric stream matters more than the others, or whether a better mapping rule improves consistency. If you add validated signal processing and a careful comparison across days, the work starts to look like real research.

Project Variations

• Use only heart rate and breathing rate to test whether a simpler input set gives more repeatable portraits.
• Replace self-reported emotion with a fixed task, such as paced breathing or mental math, and compare image stability.
• Compare portrait similarity across different art styles or prompt templates to see which setup best preserves biometric patterns.

Learn More

• NIH PubMed: Search for review articles on heart rate variability, galvanic skin response, and emotion measurement.
• NIST Open Data Portal: Look for general guidance on measurement, calibration, and reproducibility.
• NIH NCCIH: Read background pages on stress, relaxation, and physiological response.
• Understanding Digital Signal Processing by Richard G. Lyons: Find it through a library or preview edition for signal-cleaning ideas.
• MIT OpenCourseWare: Search for courses on machine learning, computer vision, and generative models.
• ImageJ documentation: Use the official docs to learn image measurement and comparison workflows.