Shadow Detection and Removal in Imaging Systems

Shadows are a pervasive artefact in optical imaging, arising whenever external illumination is occluded by objects within a scene. While often essential for conveying depth and realism, shadows can undermine image interpretation by obscuring detail, distorting spectral response and complicating tasks such as object recognition, scene reconstruction and photogrammetry. Research in shadow detection and removal spans physical modelling, statistical analysis and data-driven methods. Classical approaches exploit geometric relations between light source, surface normals and image intensity to delineate shadow boundaries, often yielding a binary shadow mask. More recent advances leverage machine-learning, notably convolutional neural networks, to distinguish shadowed from lit regions by exploiting contextual cues and chromatic invariants. Shadow removal and compensation techniques address both hard shadows and penumbra regions through methods such as spectral unmixing, illumination transfer and image inpainting. Applications of robust shadow handling include remote sensing of urban and agricultural landscapes, autonomous navigation under variable illumination, medical imaging and digital photography enhancement. The interplay between real-time performance, generality across sensors and preservation of natural appearance remains a central challenge driving ongoing research.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Shadow Detection and Removal in Imaging Systems publication trend

The graph below shows the total number of articles in shadow detection and removal in imaging systems across all publications each year (not limited to Nature Index journals).

Technical terms

Shadow mask: Binary map identifying pixels classified as shadowed or illuminated within an image.

Penumbra: Transitional region of partial shading between full illumination and full shadow, characterised by gradual intensity change.

Endmember: Pure spectral signature representing a distinct material used as a basis in spectral unmixing of hyperspectral data.

Spectral unmixing: Process of decomposing mixed pixel spectra into constituent endmember contributions and their abundances.

References

1. Shadow Detection and Compensation from Remote Sensing Images under Complex Urban Conditions. Remote Sensing (2021).
2. Shadow Detection and Restoration for Hyperspectral Images Based on Nonlinear Spectral Unmixing. Remote Sensing (2020).
3. A Moving Shadow Elimination Method Based on Fusion of Multi-Feature. IEEE Access (2020).