Infrared light-guided portrait relighting

Patent metadata

Field	Value
Publication number	US-12477232-B2
Application number	US-202117754626-A
Country	US
Kind code	B2
Filing date	Apr 8, 2021
Priority date	Nov 9, 2020
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

An imaging system includes a processor, a memory, a visible light camera configured to record a first image of a scene, and an infrared camera configured to record a second image of the scene. The processor configured to execute instructions stored in the memory to input the first image and the second image into a neural network. The neural network relights the first image, based on characteristics of the second image, to correspond to an image of the scene under canonical illumination conditions.

First claim

What is claimed is:

1 . An imaging system, comprising:
a visible light camera configured to record a first image of a scene; an infrared camera configured to record a second image of the scene; a processor; and a memory, the processor configured to execute instructions stored in the memory to input the first image and the second image into a neural network, the neural network performing image segmentation on the first image and the second image to generate a segmented first image and a segmented second image and relighting the segmented first image, based on characteristics of the segmented second image, to output an image of the scene under canonical illumination conditions, wherein:
the canonical illumination conditions include a key light, a fill light, and a back light of a three-point lighting setup, and
the neural network was trained by processing pairs of visible light and infrared images of a training scene from a set of training images through the neural network.

2 . The imaging system of claim 1 , further comprising: an infrared light source configured to illuminate the scene with infrared illumination.

3 . The imaging system of claim 2 , wherein the infrared light source is collocated with the infrared camera.

4 . The imaging system of claim 1 , wherein the visible light camera and the infrared camera are collocated and have a same or overlapping fields of view.

5 . The imaging system of claim 1 , wherein the processor is configured to operate the visible light camera and the infrared camera to record the first image and the second image simultaneously in a single exposure period.

6 . The imaging system of claim 1 , wherein the neural network is a convolutional neural network.

7 . The imaging system of claim 6 , wherein the convolutional neural network includes a U-Net.

8 . The imaging system of claim 1 , wherein the neural network relights the segmented first image by at least one of:
removing shadows; removing specular artifacts; or synthesizing specular features.

9 . A computer-implemented method, comprising:
receiving a red-green-blue, RGB, image of a scene recorded by an RGB camera in a lighting environment; obtaining an infrared image of the scene recorded by an infrared camera under infrared light illumination of the scene, inputting both the RGB image and the infrared image of the scene into a neural network, the neural network performing image segmentation on the RGB image and the infrared image to generate a segmented RGB image and a segmented infrared image and relighting the segmented RGB image based on characteristics of the segmented infrared image of the scene to correspond to a relighted RGB image obtained under canonical illumination conditions, wherein:
the canonical illumination conditions include a key light, a fill light, and a back light of a three-point lighting setup, and
the neural network was trained by processing pairs of visible light and infrared images of a training scene from a set of training images through the neural network; and
receiving the relighted RGB image as an output of neural network.

10 . The method of claim 9 , wherein the neural network relights the segmented RGB image by at least one of:
removing shadows; removing specular artifacts; or synthesizing specular features.

11 . The method of claim 9 , wherein obtaining the infrared image of the scene includes collocating the RGB camera and the infrared camera.

12 . The method of claim 11 , wherein obtaining the infrared image includes recording the infrared image simultaneously as a recording of the RGB image of the scene by the RGB camera.

13 . The method of claim 9 , wherein the neural network was trained by further:
providing a light stage including multiple light sources to controllably illuminate scenes; providing a camera system including a visible light camera and an infrared camera to capture images of the scenes under controlled illumination conditions; recording the set of training images using the camera system for training the neural network; processing the pairs of visible light and infrared images of the scene from the set of training images through the neural network to produce a relighted red-green-blue, RGB, image of the scene as an output image of the scene; and comparing the relighted RGB image for the scene with a target canonical RGB image of the scene, and adjusting neural network parameters to reduce differences between the relighted RGB image for the scene and the target canonical RGB image of the scene.

14 . A method for training a neural network to relight visible light images, the method, comprising:
providing a light stage including multiple light sources to controllably illuminate scenes; providing a camera system including a visible light camera and an infrared camera to capture images of the scenes under controlled illumination conditions; recording a set of training data images using the camera system for training the neural network, processing a pair of visible light and infrared images of a scene from the set of training data images through the neural network by performing image segmentation on the pair of visible light and infrared images to generate a segmented visible light image and a segmented infrared image and to produce a relighted red-green-blue, RGB, image of the scene as an output image of the scene; and comparing the relighted RGB image for the scene with a target canonical RGB image of the scene, and adjusting neural network parameters to reduce differences between the relighted RGB image for the scene and the target canonical RGB image of the scene, wherein the target canonical RGB image of the scene corresponds to an image of the scene obtained under canonical illumination conditions that include a key light, a fill light, and a back light of a three-point lighting setup.

15 . The method of claim 14 , wherein recording the set of training data images includes adjusting the multiple light sources in the light stage to correspond to a canonical lighting environment for the scene, and recording the target canonical RGB image of the scene.

16 . The method of claim 14 , wherein recording the set of training data images using the camera system includes turning on the multiple light sources in the light stage, one-by-one, and recording pairs of one-light-at-a-time, OLAT, RGB images and OLAT infrared images of the scene.

17 . The method of claim 16 , further comprising combining a first OLAT RGB image of a scene obtained under illumination from one light source with a second RGB image or images obtained under illumination from a second light source or light sources to simulate a combined RGB image of the scene.

18 . The method of claim 16 , further comprising combining all OLAT RGB images of a scene obtained under illumination from one light source may be combined with an RGB image or images obtained under illumination from other light sources to simulate a combined RGB image.

19 . The method of claim 14 , wherein recording the set of training data images using the camera system includes recording RGB images and infrared images of a plurality of different scenes.

20 . The method of claim 14 , wherein adjusting the neural network parameters to reduce differences between the relighted RGB image for the scene and the target canonical RGB image of the scene involves evaluating a perceptual loss function.

21 . The method of claim 14 , wherein adjusting the neural network parameters t

Assignees

• Google Llc

Inventors

• Lawrence Jason
• Achar Supreeth
• Xia Zhihao

Classifications

• H04N23/20

  for generating image signals from infrared radiation only · CPC title

• G06T5/70

  Denoising; Smoothing · CPC title

• G06T2207/10012

  Stereo images · CPC title

• H04N23/84

  for processing colour signals · CPC title

• H04N23/11 Primary

  for generating image signals from visible and infrared light wavelengths · CPC title