Color Illumination Image Separation from Single Complex Scene Image for Augmented Reality

Estimating scene illumination is crucial for seamless augmented reality integration. We present a method to extract scene color illumination from a single real-world image, enabling realistic virtual manipulations like relighting and material replacement. We extend the traditional white illumination assumption to colored illumination and use deep intrinsic decomposition to separate the scene image into colored reflectance and illumination components. After intensity correction, we perform channel separation to obtain illumination images under different spectra. By manipulating these spectral illuminations, we can create augmented reality images with realistic lighting and material effects.