Mobile Portrait Relighting With Surface Geometry And Light Models

The patent applies professional portrait lighting models (such as Rembrandt, butterfly, and loop lighting) that were traditionally implemented with complex physical equipment to a mobile device camera system. The machine learning models replicate the effects of professional lighting setups by analyzing captured images and applying appropriate lighting transformations, making professional-grade lighting effects accessible through software rather than requiring specialized hardware

The patent replaces mechanical lighting equipment (off-camera flashes, reflectors, and physical lighting modifiers) with computational methods. Machine learning models process captured images to determine surface geometry, estimate environmental lighting, calculate light energy distribution, and generate relit images that mimic professional lighting effects, substituting physical lighting systems with algorithm-based lighting synthesis

The patent implements automated lighting analysis and application through machine learning models that independently perform multiple tasks: analyzing surface geometry of the subject, estimating environmental lighting conditions, determining optimal lighting parameters, and applying appropriate lighting effects. The system serves itself by automatically processing captured images through the complete lighting enhancement pipeline without requiring manual intervention for each step

The patent pre-trains machine learning models with extensive portrait lighting data and professional lighting knowledge before deployment. The models are prepared in advance to recognize various lighting scenarios, understand surface geometry characteristics, and know the appropriate lighting corrections to apply. This preliminary training enables the system to immediately provide professional-grade lighting enhancement when capturing portraits, without requiring users to have expert knowledge