Reflectance and Illuminant Estimation
From VISTA LAB WIKI
We have been working on ways to estimate an object's reflectance under a known illuminant using a few possibly noisy measurements of the reflectance. Future work may extend into illuminant estimation and color conversion. Steve Lansel is heading up this work.
The problem of estimating a reflectance signal (at least 30 samples) from a small (3-8) number of sensor measurements is an underconstrained problem. Fortunately, reflectance functions have properties such as smoothness that can be exploited to still give reasonable estimates when this a priori information is employed. Traditional methods employ linear reconstruction based on principal components analysis (PCA) or Wiener filtering. We have developed improved algorithms for reflectance estimation by relying on nonlinear methods based upon sparse representations and local estimations, which more accurately captures the a priori information of the nature of reflectances.
 Methods of Reflectance Estimation (to describe and evaluate)
- PCA subspace
- Wiener filter
- Sparse reflectance recovery with dictionary from MOD-Recovery
- Local averaging
- Local linear regression
High level description of main routines
Details of running the code and putting it in the processing pipeline should be placed in the page of the Hyperspectral/Recovery section.
- Spatio-spectral reconstruction of the multispectral datacube using sparse recovery
- Dictionaries for sparse representation and recovery of reflectances
SVN check-in, code organization.
Visualizations of the algorithms and their choices
MOD description, put paper here, examples of figures.