ref: ba413ddffe893bcaac763e1c08920c0312a20b33
parent: 84b72ac10d75b4b93b9149ec1d3cbb9984ad533c
author: phil9 <telephil9@gmail.com>
date: Thu Dec 8 14:42:57 EST 2022
blur: code factoring box blur is like gaussian blur but uses a non-weighted convolution matrix instead so there's no use to have two functions. The kernel computation is split from the actual convolution allowing to use an hard-coded kernel for box blur and thus the same function.
--- a/blur.c
+++ b/blur.c
@@ -1,54 +1,51 @@
#include "a.h"
-
-typedef struct Filter Filter;
-struct Filter
-{- const char *name;
- filterfn filter;
-};
+enum { Fbox, Fgaussian, };+char* filters[] = { "box", "gaussian" };+double *kernel;
int size;
-uchar*
-box(uchar *data, int w, int h, int depth)
+double boxkernel[] = {+ 1./9., 1./9., 1./9.,
+ 1./9., 1./9., 1./9.,
+ 1./9., 1./9., 1./9.,
+};
+
+double*
+computekernel(int radius, double σ)
{- uchar *out;
- int n, x, y, c, s;
+ const double Π = 3.14159265358979323846;
+ double *k, d, g, s;
+ int kw, x, y;
- n = depth*w*h*sizeof(uchar);
- out = eallocbuf(n);
- memmove(out, data, n);
- /* box blur is not defined for edge points */
- for(y = 1; y < h - 1; y++){- for(x = 1; x < w - 1; x++){- for(c = 0; c < 3; c++){ /* R G B */-#define P²(X,Y,C) (data[((X) + w*(Y))*depth + C] * data[((X) + w*(Y))*depth + C])
- s = P²(x - 1, y + 1, c) +
- P²(x + 0, y + 1, c) +
- P²(x + 1, y + 1, c) +
- P²(x - 1, y + 0, c) +
- P²(x + 0, y + 0, c) +
- P²(x + 1, y + 0, c) +
- P²(x - 1, y - 1, c) +
- P²(x + 0, y - 1, c) +
- P²(x + 1, y - 1, c);
- out[(x + w * y) * depth + c] = sqrt(s / 9);
-#undef P²
- }
+ kw = 2*radius+1;
+ k = malloc(kw*kw*sizeof(double));
+ if(k == nil)
+ sysfatal("malloc: %r");+ d = 1.0/(2.0*Π*σ*σ);
+ s = 0.0;
+ for(y = -radius; y <= radius; y++){+ for(x = -radius; x <= radius; x++){+ g = d*exp(-(x*x+y*y)/(2*σ*σ));
+ k[(x+radius) + kw * (y+radius)] = g;
+ s += g;
}
}
- return out;
+ for(y = 0; y < kw; y++){+ for(x = 0; x < kw; x++){+ k[x + kw * y] /= s;
+ }
+ }
+ return k;
}
uchar*
-gaussian(uchar *data, int w, int h, int depth)
+convolve(uchar *data, int w, int h, int depth)
{- const double E = 2.7182818284590452354;
- const double Π = 3.14159265358979323846;
uchar *out;
- int kw, x, y, kx, ky;
- double *k, σ, s, e, n, d, v, r, g, b;
+ int kw, n, x, y, kx, ky;
+ double v, r, g, b;
if(size < 0)
sysfatal("gaussian filter needs a size argument");@@ -55,33 +52,13 @@
n = depth*w*h*sizeof(uchar);
out = eallocbuf(n);
memmove(out, data, n);
- σ = (double)size / 2.0;
- kw = 2 * size + 1;
- k = malloc(kw*kw*sizeof(double));
- if(k == nil)
- return nil;
- s = 0.0f;
- for(y = -size; y <= size; y++){- for(x = -size; x <= size; x++){- n = (double)(-(x * x + y * y));
- d = 2 * σ * σ;
- e = pow(E, n / d);
- v = e / (2 * Π * σ * σ);
- k[(x + size) + kw * (y + size)] = v;
- s += v;
- }
- }
- for(y = 0; y < kw; y++){- for(x = 0; x < kw; x++){- k[x + kw * y] /= s;
- }
- }
+ kw = 2*size+1;
for(y = size; y < (h - size); y++){ for(x = size; x < (w - size); x++){r = g = b = 0.0;
for(ky = -size; ky <= size; ky++){ for(kx = -size; kx <= size; kx++){- v = k[(kx + size) + kw * (ky + size)];
+ v = kernel[(kx + size) + kw * (ky + size)];
#define P(X,Y,C) data[((X) + (w*(Y)))*depth + C]
r += (double)P(x - kx, y - ky, 0) * v;
g += (double)P(x - kx, y - ky, 1) * v;
@@ -94,7 +71,6 @@
out[(x + w*y)*depth + 2] = b;
}
}
- free(k);
return out;
}
@@ -105,26 +81,21 @@
exits("usage");}
-static Filter filters[] = {- { "box", box },- { "gaussian", gaussian },-};
-
void
main(int argc, char *argv[])
{- Filter *f;
char *fname;
- int i;
+ int f, i;
+ fname = nil;
size = 2;
- f = nil;
+ f = -1;
ARGBEGIN{case 'f':
fname = EARGF(usage());
for(i=0; i<nelem(filters); i++){- if(strcmp(fname, filters[i].name)==0){- f = filters+i;
+ if(strcmp(fname, filters[i])==0){+ f = i;
break;
}
}
@@ -136,7 +107,16 @@
usage();
break;
}ARGEND;
- if(f==nil)
+ if(f<0)
usage();
- applyfilter(f->name, f->filter);
+ switch(f){+ case Fbox:
+ size = 1;
+ kernel = boxkernel;
+ break;
+ case Fgaussian:
+ kernel = computekernel(size, 1.0);
+ break;
+ }
+ applyfilter(fname, convolve);
}