Illumination recovery from image with cast shadows via sparse representation.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Recovering scene illumination from images with cast shadows is challenging due to their complexity.
• Traditional methods struggle to approximate images with cast shadows using low-dimensional linear subspaces.

Purpose of the Study:

• To propose a novel sparse representation method for recovering scene illumination from single images containing cast shadows.
• To efficiently model and represent complex shadow effects in Lambertian scenes.

Main Methods:

• Modeling images with cast shadows as a diffusive part and a residual shadow part.
• Employing an l(1)-regularized least-squares formulation with nonnegativity constraints.
• Leveraging advances in compressive sensing for an effective and fast solution.

Main Results:

• The proposed sparse representation method effectively recovers scene illumination.
• The approach demonstrates favorable performance compared to existing methods in experiments.
• Successful validation on both synthetic and real-world datasets.

Conclusions:

• Sparse representation offers an efficient and accurate solution for illumination recovery in images with cast shadows.
• The method's ability to handle complex shadow details advances the field of computational photography.
• This technique provides a robust framework for analyzing and reconstructing scene properties from single images.