A shape-adaptive structural constraint

We exploit the structural correlation by imposing that the three transforms supports which are used for the filtering of the Y, U, and V channels at a particular location have the same adaptive shape.

(In practice, we use for all three channels the adaptive neighborhoods defined by the anisotropic LPA-ICI for the luminance channel.)

Such a constraint make so that whenever some structure is detected (by the LPA-ICI), it is taken into account (and thus preserved) for the filtering of all three channels. Restricted to the adaptive supports, however, the channels are assumed as independent, and thus the transform-domain filtering is still performed for each channel independently from the others.