Related Experiment Videos
Shadow-invariant classification for scenes illuminated by daylight
1Image Analysis and Control Group, Silsoe Research Institute, Bedford, UK. john.marchant@bbsrc.ac.uk
Summary
This study introduces a physics-based shadow compensation method for daylight images. It transforms image ratios to create a shadow-independent classification, improving outdoor scene analysis.
Area of Science:
- Computer Vision
- Image Processing
- Physics-based Rendering
Background:
- Shadows in daylight scenes complicate image analysis and classification.
- Existing methods may struggle with varying illumination conditions and surface reflectivities.
Purpose of the Study:
- To develop a robust shadow compensation technique for daylight-illuminated scenes.
- To enable shadow-independent image classification by transforming image data.
Main Methods:
- A physics-based approach utilizing simplified blackbody radiation and CIE daylight models.
- Mathematical transformation of red/blue (rm) and green/blue (gm) ratios into a power law relationship.
- Pre-calculation of the exponent 'A' based on camera filter characteristics and daylight models.
Main Results:
- Demonstrated that the power law relationship holds even with finite camera bandwidths and the CIE daylight model.
- Transformed images showed similar gray-level distributions for shadowed and non-shadowed areas.
- Thresholding transformed images yielded effective shadow-independent classifications.
Conclusions:
- The proposed method provides effective shadow compensation in outdoor images.
- The transformation enables reliable image classification irrespective of shadow presence.
- This technique enhances the analysis of scenes with complex lighting.