//This function takes a noise function onto a "nearby" (in terms of function distance) noise function.
//Given the noise function, its dimension, its bounds, a uniform distribution of perturbation [-amt, amt], the fraction of points to be perturbed, and the amount of saturation to apply on renormalization.
//The exact bound is extremely difficult to calculate, but for saturation = 1, assuming fraction * number of samples is an integer, and assuming no sample is taken out of bounds, the maximum change to the integral over the unit hypercube is fraction * amt.
void perturb_noise_sum(noise_sum* n, unsigned dimension, float n0, float n1, float amt, float fraction, float saturation) {
assert(n0 <= n1);
unsigned ncount = upow(n->n.count, dimension);
unsigned perturbCt = uceilf(ncount * fraction + 1.0 - EPSILON);
for(unsigned i = 0; i < perturbCt; i++) {
unsigned idx = uniformNatural(ncount);
//This check should never be violated. However, we may want to be more tolerant in the future: if this is violated for good reasons, we may want to remove the check.
assert(n->n.values[idx] >= n0 && n->n.values[idx] <= n1);
n->n.values[idx] = n->n.values[idx] + uniformFloatS(amt);
}
//Apply some blur
blur_noise(&n->n, dimension, 0.25);
//Convert to 1d noise.
unsigned ocount = n->n.count;
n->n.count = ncount;
//Normalize
noise_rescale_out(&n->n, n0, n1);
//Saturate
noise_saturate(&n->n, saturation);
//Convert back to d dimensional noise.
n->n.count = ocount;
}
//--------------------------------------------------------------------
void image_filter_lut::realloc_lut(real radius)
{
m_radius = radius;
m_diameter = uceilf(radius) * 2;
m_start = -int(m_diameter / 2 - 1);
unsigned size = m_diameter << image_subpixel_shift;
if(size > m_weight_array.size())
{
m_weight_array.resize(size);
}
}