Portrait HDR Illumination Estimation Using Multi-BRDF Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for estimating lighting from a single image of a face struggle to accurately recover omnidirectional HDR illumination due to the inherent ambiguity between light source strength and surface albedo, particularly for diverse skin tones, and are limited by relying on single reflectance functions that do not account for complex skin reflections.

Innovation Solution

A machine learning system is trained using a dataset of portraits captured under varied lighting conditions, employing multiple bidirectional reflection distribution functions as a loss function, and utilizes a large dataset of 1 million indoor and outdoor environments to estimate HDR illumination, incorporating a multi-scale adversarial loss for robustness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If a single reflectance function is used for lighting estimation, then the method is simple to implement, but it fails to accurately recover omnidirectional HDR illumination for diverse skin tones

Engineering Contradiction:
Improveease of implementationVSAvoidlighting estimation accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent changes the reflectance function parameters by using multiple BRDFs with different reflectance properties instead of a single reflectance function. This allows the system to adapt to diverse skin tones and lighting conditions, improving measurement precision while maintaining implementation feasibility through pre-computed BRDF libraries.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent combines multiple BRDFs (bidirectional reflection distribution functions) to create a composite reflectance model. This composite approach integrates different material responses to light, enabling accurate representation of diverse skin pigments and complex reflections, thereby resolving the contradiction between simplicity and accuracy.

Inventive Principle:
Principle #40Composite materials

2Measurement precision

If multiple BRDFs are used to improve lighting estimation accuracy, then lighting recovery precision improves, but computational complexity increases

Engineering Contradiction:
Improvelighting recovery accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary computation by pre-computing and storing BRDF libraries before runtime. This allows the system to have complex multiple BRDF models ready in advance, reducing real-time computational complexity while maintaining high lighting recovery accuracy during actual execution.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses pre-computed BRDF copies stored in libraries instead of computing complex reflectance functions in real-time. This copying approach allows the system to leverage multiple BRDFs for accurate lighting estimation without the full computational burden during runtime, resolving the complexity-accuracy trade-off.

Inventive Principle:
Principle #26Copying

3Ease of manufacture

If a small database of individuals is used for training, then training data collection is simple, but the model fails to generalize to diverse skin pigments

Engineering Contradiction:
Improvedata collection simplicityVSAvoidgeneralization to diverse skin tones
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal training approach by collecting data from diverse individuals with various skin pigments, expressions, and poses. This universal dataset enables the model to generalize across different skin tones and conditions, transforming a simple data collection process into a multi-functional training resource that handles diverse scenarios.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent changes the diversity parameters of the training database by intentionally including subjects with diverse skin pigments, expressions, and poses. This parameter diversification allows the model to learn robust lighting estimation that generalizes to unseen individuals, resolving the contradiction between data collection simplicity and model versatility.

Inventive Principle:
Principle #35Parameter changes

4Productivity

If real-time processing is required for AR applications, then user experience is improved, but lighting estimation accuracy may be compromised

Engineering Contradiction:
Improveprocessing speedVSAvoidlighting estimation accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary computation by pre-computing BRDF libraries and training the neural network model in advance. This allows the system to achieve real-time processing speeds during AR execution while maintaining high lighting estimation accuracy, as the complex computations are completed beforehand.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses pre-computed BRDF copies and a trained neural network model that can be rapidly applied in real-time. This copying approach enables the system to leverage accurate lighting estimation capabilities without performing complex computations during real-time AR processing, resolving the speed-accuracy trade-off.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3939011B1Learning illumination from diverse portraits
Publication Date: 2026.01.28 GOOGLE LLC
  • EP3939011B1 patent drawingFigure 1
  • EP3939011B1 patent drawingFigure 2
  • EP3939011B1 patent drawingFigure 3

AI summary

Techniques of estimating lighting from portraits includes generating a lighting estimate from a single image of a face based on a machine learning (ML) system using multiple bidirectional reflection distribution functions (BRDFs) as a loss function. In some implementations, the ML system is trained using images of faces formed with HDR illumination computed from LDR imagery. The technical solution includes training a lighting estimation model in a supervised manner using a dataset of portraits and their corresponding ground truth illumination.