Laplace-distributed increments, the Laplace prior, and edge-preserving regularization

摘要: For a given two-dimensional image, we define the horizontal and vertical increments at pixel location to be difference between intensity values that neighboring pixels right above, respectively. typical it makes intuitive sense will usually near zero, corresponding areas of smooth variation in image intensity, but often have large magnitude, edges where sharp changes occur. In this paper, explore use Laplace increment model, which are assumed independent identically distributed random variables – distribution with heavy tails allowing for zero mean. The prior constructed from model is very similar total (TV) prior. We perform theoretical analysis its properties, shows yields regularization scheme regularized solutions contained space bounded variation, just as TV Moreover, numerical experiments indicate reconstructions qualitatively those obtained using TV.