Simulated Reflectance

Author

Tina Lasisi

Published

April 1, 2026

1 Overview of Simulated Skin Reflectance Datasets

This analysis examines three distinct datasets of simulated skin reflectance spectra:

  1. Training Dataset: Wide parameter ranges for model training
  2. Test Dataset (Wide): Same parameter ranges as training for validation
  3. Test Dataset (Narrow): Restricted parameter ranges to test model performance on a specific subset

Note: All datasets measure reflectance from 380-1000nm wavelength range, though plots focus on the visible spectrum (400-700nm).

Terminology: “Parameter range” refers to the minimum and maximum values allowed for each skin property in the simulation. For example, thickness range 0.000-1.000 means skin thickness can vary anywhere from 0 to 1 (in appropriate units).

1.1 Dataset Parameters

1.1.1 Parameters that remain CONSTANT across all datasets:

  • Melanin (ME): 0.001-0.430 - Total melanin content controlling overall skin darkness
  • Blood Volume (BL): 0.000-0.200 - Amount of blood in skin affecting redness/pinkness
  • Oxygenation (OX): 0.900-1.000 - Blood oxygen saturation level (high = red, low = blue)

1.1.2 Parameters that VARY between datasets:

  • Thickness (TH): Physical thickness of skin layer affecting how light scatters
    • Wide range: 0.000-1.000 (full physiological range)
    • Narrow range: 0.001-0.035 (very thin skin only)
  • Eumelanin (EU): Fraction of melanin that is eumelanin (brown/black) vs pheomelanin (red/yellow)
    • Wide range: 0.650-0.800 (moderate to high eumelanin)
    • Narrow range: 0.700-0.780 (high eumelanin only)

1.2 Meta Spectrum Overview

First, let’s examine the overall meta spectrum file that contains the full range of simulated spectra:

Sample of skin reflectance spectra colored by melanin index

2 Comparison of Three Datasets

2.1 Summary of Dataset Differences

Only TWO parameters vary between datasets: 1. Thickness (TH): Skin thickness affecting light scattering 2. Eumelanin (EU): Ratio of brown/black vs red/yellow pigment

All other parameters (Melanin, Blood Volume, Oxygenation) remain constant across all three datasets.

“Narrow” refers to the parameter RANGES being narrower, not the skin itself.

2.2 Dataset 1: Training Data with Wide Parameter Ranges

  • Purpose: Train models on the full range of skin thickness and eumelanin values
  • Thickness: 0.000-1.000 (full range from very thin to thick skin)
  • Eumelanin: 0.650-0.800 (moderate to high brown/black pigment)

2.3 Dataset 2: Test Data with Wide Parameter Ranges

  • Purpose: Test model performance on same parameter space as training
  • Thickness: 0.000-1.000 (matches training data)
  • Eumelanin: 0.650-0.800 (matches training data)

2.4 Dataset 3: Test Data with Narrow Parameter Ranges

  • Purpose: Evaluate model on a specific subset - very thin skin with high eumelanin
  • Thickness: 0.001-0.035 (only very thin skin - ~3.5% of the full range)
  • Eumelanin: 0.700-0.780 (high eumelanin only - ~53% of the training range)
  • “Narrow” refers to: Both thickness AND eumelanin parameter RANGES being narrower/restricted compared to training data
  • Physical interpretation: This represents very thin skin (like eyelids) with predominantly brown/black pigmentation

2.5 Side-by-Side Comparison of All Three Datasets

2.5.1 Complete Parameter Values for Each Dataset

Parameter ranges for all three datasets. Bold values indicate parameters that vary between datasets.
Parameter Training Dataset Test Dataset (Wide) Test Dataset (Narrow)
Melanin (ME) 0.001-0.430 0.001-0.430 0.001-0.430
Blood Volume (BL) 0.000-0.200 0.000-0.200 0.000-0.200
Thickness (TH) 0.000-1.000 0.000-1.000 0.001-0.035
Eumelanin (EU) 0.650-0.800 0.650-0.800 0.700-0.780
Oxygenation (OX) 0.900-1.000 0.900-1.000 0.900-1.000
NoteKey differences
  • Training & Test (Wide): Identical parameters across the full physiological range.
  • Test (Narrow): Restricted ranges for both thickness (0.001-0.035) and eumelanin (0.700-0.780).
  • The narrow dataset tests whether models trained on broad ranges can generalize to a constrained subspace.
Per-dataset parameter summary.
Dataset ME (Melanin) BL (Blood) TH (Thickness) EU (Eumelanin) OX (Oxygenation)
Training 0.001-0.430 0.000-0.200 0.000-1.000 0.650-0.800 0.900-1.000
Test (Wide) 0.001-0.430 0.000-0.200 0.000-1.000 0.650-0.800 0.900-1.000
Test (Narrow) 0.001-0.430 0.000-0.200 0.001-0.035 0.700-0.780 0.900-1.000
Range compression in the narrow test dataset.
Parameter Training/Wide Range Narrow Range Narrow as % of Wide
Thickness (TH) 1.000 0.034 3.4%
Eumelanin (EU) 0.150 0.080 53.3% 
THICKNESS (TH)
Training/Wide: |==================================| 0.000 to 1.000
Narrow:        |=|                                  0.001 to 0.035

EUMELANIN (EU)
Training/Wide: |==================================| 0.650 to 0.800
Narrow:        |       ================|            0.700 to 0.780

2.6 Key Observations

2.6.1 Training vs Test (Wide) Datasets

  • Both contain the same parameter ranges
  • This allows for direct validation of model performance
  • Any differences in model performance can be attributed to the specific spectra sampled, not parameter range differences

2.6.2 Test (Narrow) Dataset Characteristics

  • “Narrow” means BOTH parameters are restricted:
    • Thickness reduced from 0.000-1.000 to 0.001-0.035 (only ~3.5% of original range)
    • Eumelanin reduced from 0.650-0.800 to 0.700-0.780 (only ~53% of original range)
  • Focuses exclusively on very thin skin with high eumelanin content
  • Tests whether models trained on wide ranges can accurately predict spectra for this specific physiological subset
  • Represents a challenging test case as it uses only a small portion of the training parameter space

2.7 Implications for Model Development

  1. Generalization Testing: The wide test set validates general model performance across the full parameter space
  2. Specific Case Testing: The narrow test set evaluates performance on a challenging subset (very thin, high-eumelanin skin)
  3. Parameter Coverage:
    • Training covers the full physiological range
    • Test (Narrow) uses only ~3.5% of the thickness range and ~53% of the eumelanin range
    • This tests model interpolation capability within a restricted parameter space