The restored image is predicted from a corrupted observation after training on a set of sample images .
A shrinkage (mapping) function is directly modeled as a linear combination of radial basis function kernels, where is the shared precision parameter, denotes the (equidistant) kernel positions, and M is the number of Gaussian kernels.
Because the shrinkage function is directly modeled, the optimization procedure is reduced to a single quadratic minimization per iteration, denoted as the prediction of a shrinkage field where denotes the discrete Fourier transform and is the 2D convolution with point spread function filter, is an optical transfer function defined as the discrete Fourier transform of , and is the complex conjugate of .
is learned as for each iteration with the initial case , this forms a cascade of Gaussian conditional random fields (or cascade of shrinkage fields (CSF)). Loss-minimization is used to learn the model parameters .
The learning objective function is defined as , where is a differentiableloss function which is greedily minimized using training data and .
Performance
Preliminary tests by the author suggest that RTF5[1] obtains slightly better denoising performance than , followed by , , , and BM3D.
BM3D denoising speed falls between that of and , RTF being an order of magnitude slower.
Advantages
Results are comparable to those obtained by BM3D (reference in state of the art denoising since its inception in 2007)
Minimal runtime compared to other high-performance methods (potentially applicable within embedded devices)
Parallelizable (e.g.: possible GPU implementation)
Predictability: runtime where is the number of pixels