Hello all? i’m glad to be part of the processing community. and i’ll try to help my best.
I’ve got recent issues using the noise function. the precision was not excellent.
for instance a simple curve taken from the noise function gives me some strange staircase forms
here is an export svg…
I’ve tried different approaches, like create svg in very big definition files, I imagines that it was round to pixels troubles… but it was none of that the export svg process of processing is quite good and there is no rounding trouble.
the fact was that the noise fucntion in processing (java) uses a cosinus table (cosLUT) whose definition is not excellent.
taking the code from the open source page of processing and rewrite is just by changing how the cosinus table was made changed the definition.
Here is the full version of my noise.pde
import java.util.Random;
//////////////////////////////////////////////////////////////
// PERLIN NOISE, rewriten by cfloutier to change the way cosinus is computed
// [toxi 040903]
// octaves and amplitude amount per octave are now user controlled
// via the noiseDetail() function.
// [toxi 030902]
// cleaned up code and now using bagel's cosine table to speed up
// [toxi 030901]
// implementation by the german demo group farbrausch
// as used in their demo "art": http://www.farb-rausch.de/fr010src.zip
static final int PERLIN_YWRAPB = 4;
static final int PERLIN_YWRAP = 1<<PERLIN_YWRAPB;
static final int PERLIN_ZWRAPB = 8;
static final int PERLIN_ZWRAP = 1<<PERLIN_ZWRAPB;
static final int PERLIN_SIZE = 4095;
int perlin_octaves = 4; // default to medium smooth
float perlin_amp_falloff = 0.5f; // 50% reduction/octave
// [toxi 031112]
// new vars needed due to recent change of cos table in PGraphics
int perlin_PI;
static final int perlin_TWOPI = 65536;
float[] perlin_cosTable;
float[] perlin;
Random perlinRandom;
/**
*/
public float my_noise(float x) {
// is this legit? it's a dumb way to do it (but repair it later)
return my_noise(x, 0f, 0f);
}
/**
*/
public float my_noise(float x, float y) {
return my_noise(x, y, 0f);
}
/**
* ( begin auto-generated from noise.xml )
*
* Returns the Perlin noise value at specified coordinates. Perlin noise is
* a random sequence generator producing a more natural ordered, harmonic
* succession of numbers compared to the standard <b>random()</b> function.
* It was invented by Ken Perlin in the 1980s and been used since in
* graphical applications to produce procedural textures, natural motion,
* shapes, terrains etc.<br /><br /> The main difference to the
* <b>random()</b> function is that Perlin noise is defined in an infinite
* n-dimensional space where each pair of coordinates corresponds to a
* fixed semi-random value (fixed only for the lifespan of the program).
* The resulting value will always be between 0.0 and 1.0. Processing can
* compute 1D, 2D and 3D noise, depending on the number of coordinates
* given. The noise value can be animated by moving through the noise space
* as demonstrated in the example above. The 2nd and 3rd dimension can also
* be interpreted as time.<br /><br />The actual noise is structured
* similar to an audio signal, in respect to the function's use of
* frequencies. Similar to the concept of harmonics in physics, perlin
* noise is computed over several octaves which are added together for the
* final result. <br /><br />Another way to adjust the character of the
* resulting sequence is the scale of the input coordinates. As the
* function works within an infinite space the value of the coordinates
* doesn't matter as such, only the distance between successive coordinates
* does (eg. when using <b>noise()</b> within a loop). As a general rule
* the smaller the difference between coordinates, the smoother the
* resulting noise sequence will be. Steps of 0.005-0.03 work best for most
* applications, but this will differ depending on use.
*
* ( end auto-generated )
*
* @webref math:random
* @param x x-coordinate in noise space
* @param y y-coordinate in noise space
* @param z z-coordinate in noise space
* @see PApplet#noiseSeed(long)
* @see PApplet#noiseDetail(int, float)
* @see PApplet#random(float,float)
*/
public float my_noise(float x, float y, float z) {
if (perlin == null) {
if (perlinRandom == null) {
perlinRandom = new Random();
}
perlin = new float[PERLIN_SIZE + 1];
for (int i = 0; i < PERLIN_SIZE + 1; i++) {
perlin[i] = perlinRandom.nextFloat(); //(float)Math.random();
}
perlin_cosTable = new float[perlin_TWOPI];
perlin_PI = perlin_TWOPI >> 1;
for (int i = 0; i < perlin_TWOPI; i++)
{
float angle = (float)i/perlin_PI;
perlin_cosTable[i] = cos( angle * PI );
}
}
if (x<0) x=-x;
if (y<0) y=-y;
if (z<0) z=-z;
int xi=(int)x, yi=(int)y, zi=(int)z;
float xf = x - xi;
float yf = y - yi;
float zf = z - zi;
float rxf, ryf;
float r=0;
float ampl=0.5f;
float n1, n2, n3;
for (int i=0; i<perlin_octaves; i++) {
int of=xi+(yi<<PERLIN_YWRAPB)+(zi<<PERLIN_ZWRAPB);
rxf=my_noise_fsc(xf);
ryf=my_noise_fsc(yf);
n1 = perlin[of&PERLIN_SIZE];
n1 += rxf*(perlin[(of+1)&PERLIN_SIZE]-n1);
n2 = perlin[(of+PERLIN_YWRAP)&PERLIN_SIZE];
n2 += rxf*(perlin[(of+PERLIN_YWRAP+1)&PERLIN_SIZE]-n2);
n1 += ryf*(n2-n1);
of += PERLIN_ZWRAP;
n2 = perlin[of&PERLIN_SIZE];
n2 += rxf*(perlin[(of+1)&PERLIN_SIZE]-n2);
n3 = perlin[(of+PERLIN_YWRAP)&PERLIN_SIZE];
n3 += rxf*(perlin[(of+PERLIN_YWRAP+1)&PERLIN_SIZE]-n3);
n2 += ryf*(n3-n2);
n1 += my_noise_fsc(zf)*(n2-n1);
r += n1*ampl;
ampl *= perlin_amp_falloff;
xi<<=1;
xf*=2;
yi<<=1;
yf*=2;
zi<<=1;
zf*=2;
if (xf>=1.0f) {
xi++;
xf--;
}
if (yf>=1.0f) {
yi++;
yf--;
}
if (zf>=1.0f) {
zi++;
zf--;
}
}
return r;
}
// [toxi 031112]
// now adjusts to the size of the cosLUT used via
// the new variables, defined above
private float my_noise_fsc(float i) {
// using bagel's cosine table instead
int index = (int)(i*perlin_PI)%perlin_TWOPI;
float cosvalue = perlin_cosTable[index];
// for a very precise cosinus, just un-comment this line (all cos table is uselless then you can remove it as well
// cosvalue = cos(i*PI);
return 0.5f*(1.0f-cosvalue);
}
// [toxi 040903]
// make perlin noise quality user controlled to allow
// for different levels of detail. lower values will produce
// smoother results as higher octaves are surpressed
/**
* ( begin auto-generated from noiseDetail.xml )
*
* Adjusts the character and level of detail produced by the Perlin noise
* function. Similar to harmonics in physics, noise is computed over
* several octaves. Lower octaves contribute more to the output signal and
* as such define the overal intensity of the noise, whereas higher octaves
* create finer grained details in the noise sequence. By default, noise is
* computed over 4 octaves with each octave contributing exactly half than
* its predecessor, starting at 50% strength for the 1st octave. This
* falloff amount can be changed by adding an additional function
* parameter. Eg. a falloff factor of 0.75 means each octave will now have
* 75% impact (25% less) of the previous lower octave. Any value between
* 0.0 and 1.0 is valid, however note that values greater than 0.5 might
* result in greater than 1.0 values returned by <b>noise()</b>.<br /><br
* />By changing these parameters, the signal created by the <b>noise()</b>
* function can be adapted to fit very specific needs and characteristics.
*
* ( end auto-generated )
* @webref math:random
* @param lod number of octaves to be used by the noise
* @see PApplet#noise(float, float, float)
*/
public void my_noiseDetail(int lod) {
if (lod>0) perlin_octaves=lod;
}
/**
* @see #noiseDetail(int)
* @param falloff falloff factor for each octave
*/
public void my_noiseDetail(int lod, float falloff) {
if (lod>0) perlin_octaves=lod;
if (falloff>0) perlin_amp_falloff=falloff;
}
/**
* ( begin auto-generated from noiseSeed.xml )
*
* Sets the seed value for <b>noise()</b>. By default, <b>noise()</b>
* produces different results each time the program is run. Set the
* <b>value</b> parameter to a constant to return the same pseudo-random
* numbers each time the software is run.
*
* ( end auto-generated )
* @webref math:random
* @param seed seed value
* @see PApplet#noise(float, float, float)
* @see PApplet#noiseDetail(int, float)
* @see PApplet#random(float,float)
* @see PApplet#randomSeed(long)
*/
public void my_noiseSeed(long seed) {
if (perlinRandom == null) perlinRandom = new Random();
perlinRandom.setSeed(seed);
// force table reset after changing the random number seed [0122]
perlin = null;
}
the definition of the cosinus table can be adjusted with this line
static final int perlin_TWOPI = 65536;
and to avoid using a cos-table just replace in my_noise_fsc
float cosvalue = perlin_cosTable[index];
by
float cosvalue = = cos(i*PI);
I hope it can help some of us that like me need good precision in exported file.
I use a large 2D plotter and this precision can be seen on paper…
Christophe