LAUSR

When photographing, we strive to reproduce the scene as perceived by human visual system, including its authentic dynamic range as well as accurate color. In image signal process pipeline, these reproductions are achieved through modules such as high dynamic range imaging and color constancy. In this work, we propose a unified color constancy framework applicable to single-exposure imaging and well-suited to the multi-exposure fusion-based HDR imaging, enabling accurate color reproduction. It is hypothesized that most natural images contain neutral areas. Through detection and further extension of these neutral areas, illuminant information can be effectively extracted. Experimental results across benchmark datasets and our established multi-exposure dataset demonstrate that the proposed method surpasses the majority of state-of-the-art statistical-based color constancy methods. To enhance the consistency with human visual perception, a compact chromatic adaptation transform is conducted, as the second stage of color constancy, following the illuminant estimation.