Adaptive Regularization Term Research Articles

Adaptive total variation denoising based on difference curvature

Image denoising methods based on gradient dependent regularizers such as Rudin et al.’s total variation (TV) model often suffer the staircase effect and the loss of fine details. In order to overcome such drawbacks, this paper presents an adaptive total variation method based on a new edge indicator, named difference curvature, which can effectively distinguish between edges and ramps. With adaptive regularization and fidelity terms, the new model has the following properties: at object edges, the regularization term is approximate to the TV norm in order to preserve the edges, and the weight of the fidelity term is large in order to preserve details; in flat and ramp regions, the regularization term is approximate to the L2 norm in order to avoid the staircase effect, and the weight of the fidelity term is small in order to strongly remove the noise. Comparative results on both synthetic and natural images demonstrate that the new method can avoid the staircase effect and better preserve fine details.

A segmentation-based regularization term for image deconvolution

This paper proposes a new and original inhomogeneous restoration (deconvolution) model under the Bayesian framework for observed images degraded by space-invariant blur and additive Gaussian noise. In this model, regularization is achieved during the iterative restoration process with a segmentation-based a priori term. This adaptive edge-preserving regularization term applies a local smoothness constraint to pre-estimated constant-valued regions of the target image. These constant-valued regions (the segmentation map) of the target image are obtained from a preliminary Wiener deconvolution estimate. In order to estimate reliable segmentation maps, we have also adopted a Bayesian Markovian framework in which the regularized segmentations are estimated in the maximum a posteriori (MAP) sense with the joint use of local Potts prior and appropriate Gaussian conditional luminance distributions. In order to make these segmentations unsupervised, these likelihood distributions are estimated in the maximum likelihood sense. To compute the MAP estimate associated to the restoration, we use a simple steepest descent procedure resulting in an efficient iterative process converging to a globally optimal restoration. The experiments reported in this paper demonstrate that the discussed method performs competitively and sometimes better than the best existing state-of-the-art methods in benchmark tests.

Reconstruction of electron spectra from depth doses with adaptive regularization

Electron spectral reconstruction of medical accelerators from measured depth doses is a practical method for providing the input initial phase space distribution at the patient surface that is required by Monte Carlo treatment planning systems. The posed inverse problem of spectral reconstruction is ill conditioned and this may lead to nonphysical oscillations in the reconstructed spectra. Use of a variational method of solution with a regularization technique removes the oscillations but tends to smooth the sharp (deltalike) energy peak that is a common feature in electron spectra generated by medical accelerators. Because the sharp peak contains a large percentage of the electrons in the spectrum, an accurate estimate of the peak width, height and position is critical to the success of the technique for spectrum reconstruction with regularization. We propose use of an adaptive regularization term as a special form of the general Tichonov regularization function. The variational method with the adaptive regularization term is applied to reconstruct electron spectra for the 6, 9, and 18 MeV electron beams of a Varian Clinac 2100C accelerator and proves to be a very simple, effective and accurate approach. Results using this variational method with adaptive regularization almost perfectly reconstruct electron spectra from depth dose distributions.