Discrete Hessian-norm denoising
= Discrete Hessian-norm denoising
{title2=$C=H^*B_\alpha$}
Replacing a discrete <gradient> by a second-difference operator $H:X\to X^4$ gives the global-norm penalty $\alpha\|Hu\|_2$. Its reconstruction is $g-P_Cg$ for $C=H^*B_\alpha$. Components in $\ker H$ are preserved; affine-image membership of this kernel depends on the boundary stencil. Pixelwise Hessian penalties instead use a product of four-component balls.