Model-based radiometric restoration

Radiance restoration via linear filtering is problematic for small targets, and estimating the thermally emitted radiance of sub-pixel targets is very difficult, resulting in large errors. However, with a priori knowledge of the geometry of a scene, a site model can be constructed to aid in radiance estimation. The scene is modeled as a linear combination of known shapes and the radiance of these shapes is adjusted to fit the observed image. This method of estimating the radiances of small to sub-pixel targets can yield significantly lower errors than linear filtering. The performance of the method on synthetic images is discussed as a function of the size, radiance, and background of the target and of the noise in the image. The model-based approach is found to outperform a simple linear filter for very small targets (diameter of a few pixels) with high contrast relative to the image noise.