Workshop o mikrokontrolérech na SKSP 2024.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

325 lines
10 KiB

2 months ago
/*
* Image Library -- Computation of image signatures
*
* (c) 2006 Pavel Charvat <pchar@ucw.cz>
*
* This software may be freely distributed and used according to the terms
* of the GNU Lesser General Public License.
*/
#undef LOCAL_DEBUG
#include <ucw/lib.h>
#include <ucw/fastbuf.h>
#include <ucw/conf.h>
#include <images/images.h>
#include <images/math.h>
#include <images/error.h>
#include <images/color.h>
#include <images/signature.h>
#include <alloca.h>
#include <math.h>
int
image_sig_init(struct image_context *ctx, struct image_sig_data *data, struct image *image)
{
ASSERT((image->flags & IMAGE_PIXEL_FORMAT) == COLOR_SPACE_RGB);
data->image = image;
data->flags = 0;
data->cols = (image->cols + 3) >> 2;
data->rows = (image->rows + 3) >> 2;
data->full_cols = image->cols >> 2;
data->full_rows = image->rows >> 2;
data->blocks_count = data->cols * data->rows;
if (data->blocks_count >= 0x10000)
{
IMAGE_ERROR(ctx, IMAGE_ERROR_INVALID_DIMENSIONS, "Image too large for implemented signature algorithm.");
return 0;
}
data->blocks = xmalloc(data->blocks_count * sizeof(struct image_sig_block));
data->area = image->cols * image->rows;
DBG("Computing signature for image of %ux%u pixels (%ux%u blocks)",
image->cols, image->rows, data->cols, data->rows);
return 1;
}
void
image_sig_preprocess(struct image_sig_data *data)
{
struct image *image = data->image;
struct image_sig_block *block = data->blocks;
uint sum[IMAGE_VEC_F];
bzero(sum, sizeof(sum));
/* Every block of 4x4 pixels */
byte *row_start = image->pixels;
for (uint block_y = 0; block_y < data->rows; block_y++, row_start += image->row_size * 4)
{
byte *p = row_start;
for (uint block_x = 0; block_x < data->cols; block_x++, p += 12, block++)
{
int t[16], s[16], *tp = t;
block->x = block_x;
block->y = block_y;
/* Convert pixels to Luv color space and compute average coefficients */
uint l_sum = 0, u_sum = 0, v_sum = 0;
byte *p2 = p;
if (block_x < data->full_cols && block_y < data->full_rows)
{
for (uint y = 0; y < 4; y++, p2 += image->row_size - 12)
for (uint x = 0; x < 4; x++, p2 += 3)
{
byte luv[3];
srgb_to_luv_pixel(luv, p2);
l_sum += *tp++ = luv[0] / 4;
u_sum += luv[1];
v_sum += luv[2];
}
block->area = 16;
sum[0] += l_sum;
sum[1] += u_sum;
sum[2] += v_sum;
block->v[0] = (l_sum >> 4);
block->v[1] = (u_sum >> 4);
block->v[2] = (v_sum >> 4);
}
/* Incomplete square near the edge */
else
{
uint x, y;
uint square_cols = (block_x < data->full_cols) ? 4 : image->cols & 3;
uint square_rows = (block_y < data->full_rows) ? 4 : image->rows & 3;
for (y = 0; y < square_rows; y++, p2 += image->row_size)
{
byte *p3 = p2;
for (x = 0; x < square_cols; x++, p3 += 3)
{
byte luv[3];
srgb_to_luv_pixel(luv, p3);
l_sum += *tp++ = luv[0] / 4;
u_sum += luv[1];
v_sum += luv[2];
}
for (; x < 4; x++)
{
*tp = tp[-(int)square_cols];
tp++;
}
}
for (; y < 4; y++)
for (x = 0; x < 4; x++)
{
*tp = tp[-(int)square_rows * 4];
tp++;
}
block->area = square_cols * square_rows;
uint inv = 0x10000 / block->area;
sum[0] += l_sum;
sum[1] += u_sum;
sum[2] += v_sum;
block->v[0] = (l_sum * inv) >> 16;
block->v[1] = (u_sum * inv) >> 16;
block->v[2] = (v_sum * inv) >> 16;
}
/* Apply Daubechies wavelet transformation */
# define DAUB_0 31651 /* (1 + sqrt 3) / (4 * sqrt 2) * 0x10000 */
# define DAUB_1 54822 /* (3 + sqrt 3) / (4 * sqrt 2) * 0x10000 */
# define DAUB_2 14689 /* (3 - sqrt 3) / (4 * sqrt 2) * 0x10000 */
# define DAUB_3 -8481 /* (1 - sqrt 3) / (4 * sqrt 2) * 0x10000 */
/* ... to the rows */
uint i;
for (i = 0; i < 16; i += 4)
{
s[i + 0] = (DAUB_0 * t[i + 2] + DAUB_1 * t[i + 3] + DAUB_2 * t[i + 0] + DAUB_3 * t[i + 1]) / 0x10000;
s[i + 1] = (DAUB_0 * t[i + 0] + DAUB_1 * t[i + 1] + DAUB_2 * t[i + 2] + DAUB_3 * t[i + 3]) / 0x10000;
s[i + 2] = (DAUB_3 * t[i + 2] - DAUB_2 * t[i + 3] + DAUB_1 * t[i + 0] - DAUB_0 * t[i + 1]) / 0x10000;
s[i + 3] = (DAUB_3 * t[i + 0] - DAUB_2 * t[i + 1] + DAUB_1 * t[i + 2] - DAUB_0 * t[i + 3]) / 0x10000;
}
/* ... and to the columns... skip LL band */
for (i = 0; i < 2; i++)
{
t[i + 8] = (DAUB_3 * s[i + 8] - DAUB_2 * s[i +12] + DAUB_1 * s[i + 0] - DAUB_0 * s[i + 4]) / 0x10000;
t[i +12] = (DAUB_3 * s[i + 0] - DAUB_2 * s[i + 4] + DAUB_1 * s[i + 8] - DAUB_0 * s[i +12]) / 0x10000;
}
for (; i < 4; i++)
{
t[i + 0] = (DAUB_0 * s[i + 8] + DAUB_1 * s[i +12] + DAUB_2 * s[i + 0] + DAUB_3 * s[i + 4]) / 0x10000;
t[i + 4] = (DAUB_0 * s[i + 0] + DAUB_1 * s[i + 4] + DAUB_2 * s[i + 8] + DAUB_3 * s[i +12]) / 0x10000;
t[i + 8] = (DAUB_3 * s[i + 8] - DAUB_2 * s[i +12] + DAUB_1 * s[i + 0] - DAUB_0 * s[i + 4]) / 0x10000;
t[i +12] = (DAUB_3 * s[i + 0] - DAUB_2 * s[i + 4] + DAUB_1 * s[i + 8] - DAUB_0 * s[i +12]) / 0x10000;
}
/* Extract energies in LH, HL and HH bands */
block->v[3] = fast_sqrt_u32(isqr(t[8]) + isqr(t[9]) + isqr(t[12]) + isqr(t[13]));
block->v[4] = fast_sqrt_u32(isqr(t[2]) + isqr(t[3]) + isqr(t[6]) + isqr(t[7]));
block->v[5] = fast_sqrt_u32(isqr(t[10]) + isqr(t[11]) + isqr(t[14]) + isqr(t[15]));
sum[3] += block->v[3] * block->area;
sum[4] += block->v[4] * block->area;
sum[5] += block->v[5] * block->area;
}
}
/* Compute featrures average */
uint inv = 0xffffffffU / data->area;
for (uint i = 0; i < IMAGE_VEC_F; i++)
data->f[i] = ((u64)sum[i] * inv) >> 32;
if (image->cols < image_sig_min_width || image->rows < image_sig_min_height)
{
data->valid = 0;
data->regions_count = 0;
}
else
data->valid = 1;
}
void
image_sig_finish(struct image_sig_data *data, struct image_signature *sig)
{
for (uint i = 0; i < IMAGE_VEC_F; i++)
sig->vec.f[i] = data->f[i];
sig->len = data->regions_count;
sig->flags = data->flags;
if (!sig->len)
return;
/* For each region */
u64 w_total = 0;
uint w_border = MIN(data->cols, data->rows) * image_sig_border_size;
int w_mul = w_border ? image_sig_border_bonus * 256 / (int)w_border : 0;
for (uint i = 0; i < sig->len; i++)
{
struct image_sig_region *r = data->regions + i;
DBG("Processing region %u: count=%u", i, r->count);
ASSERT(r->count);
/* Copy texture properties */
sig->reg[i].f[0] = r->a[0];
sig->reg[i].f[1] = r->a[1];
sig->reg[i].f[2] = r->a[2];
sig->reg[i].f[3] = r->a[3];
sig->reg[i].f[4] = r->a[4];
sig->reg[i].f[5] = r->a[5];
/* Compute coordinates centroid and region weight */
u64 x_sum = 0, y_sum = 0, w_sum = 0;
for (struct image_sig_block *b = r->blocks; b; b = b->next)
{
x_sum += b->x;
y_sum += b->y;
uint d = b->x;
d = MIN(d, b->y);
d = MIN(d, data->cols - b->x - 1);
d = MIN(d, data->rows - b->y - 1);
if (d >= w_border)
w_sum += 128;
else
w_sum += 128 + (int)(w_border - d) * w_mul / 256;
}
w_total += w_sum;
r->w_sum = w_sum;
uint x_avg = x_sum / r->count;
uint y_avg = y_sum / r->count;
DBG(" centroid=(%u %u)", x_avg, y_avg);
/* Compute normalized inertia */
u64 sum1 = 0, sum2 = 0, sum3 = 0;
for (struct image_sig_block *b = r->blocks; b; b = b->next)
{
uint inc2 = isqr(x_avg - b->x) + isqr(y_avg - b->y);
uint inc1 = fast_sqrt_u32(inc2);
sum1 += inc1;
sum2 += inc2;
sum3 += inc1 * inc2;
}
sig->reg[i].h[0] = CLAMP(image_sig_inertia_scale[0] * sum1 * ((3 * M_PI * M_PI) / 2) * pow(r->count, -1.5), 0, 255);
sig->reg[i].h[1] = CLAMP(image_sig_inertia_scale[1] * sum2 * ((4 * M_PI * M_PI * M_PI) / 2) / ((u64)r->count * r->count), 0, 255);
sig->reg[i].h[2] = CLAMP(image_sig_inertia_scale[2] * sum3 * ((5 * M_PI * M_PI * M_PI * M_PI) / 2) * pow(r->count, -2.5), 0, 255);
sig->reg[i].h[3] = (uint)x_avg * 127 / data->cols;
sig->reg[i].h[4] = (uint)y_avg * 127 / data->rows;
}
/* Compute average differences */
u64 df = 0, dh = 0;
if (sig->len < 2)
{
sig->df = 1;
sig->dh = 1;
}
else
{
uint cnt = 0;
for (uint i = 0; i < sig->len; i++)
for (uint j = i + 1; j < sig->len; j++)
{
uint d = 0;
for (uint k = 0; k < IMAGE_REG_F; k++)
d += image_sig_cmp_features_weights[k] * isqr(sig->reg[i].f[k] - sig->reg[j].f[k]);
df += fast_sqrt_u32(d);
d = 0;
for (uint k = 0; k < IMAGE_REG_H; k++)
d += image_sig_cmp_features_weights[k + IMAGE_REG_F] * isqr(sig->reg[i].h[k] - sig->reg[j].h[k]);
dh += fast_sqrt_u32(d);
cnt++;
}
sig->df = CLAMP(df / cnt, 1, 0xffff);
sig->dh = CLAMP(dh / cnt, 1, 0xffff);
}
DBG("Average regions difs: df=%u dh=%u", sig->df, sig->dh);
/* Compute normalized weights */
uint wa = 128, wb = 128;
for (uint i = sig->len; --i > 0; )
{
struct image_sig_region *r = data->regions + i;
wa -= sig->reg[i].wa = CLAMP(r->count * 128 / data->blocks_count, 1, (int)(wa - i));
wb -= sig->reg[i].wb = CLAMP(r->w_sum * 128 / w_total, 1, (int)(wb - i));
}
sig->reg[0].wa = wa;
sig->reg[0].wb = wb;
/* Store image dimensions */
sig->cols = data->image->cols;
sig->rows = data->image->rows;
/* Dump regions features */
#ifdef LOCAL_DEBUG
for (uint i = 0; i < sig->len; i++)
{
byte buf[IMAGE_REGION_DUMP_MAX];
image_region_dump(buf, sig->reg + i);
DBG("region %u: features=%s", i, buf);
}
#endif
}
void
image_sig_cleanup(struct image_sig_data *data)
{
xfree(data->blocks);
}
int
compute_image_signature(struct image_context *ctx, struct image_signature *sig, struct image *image)
{
struct image_sig_data data;
if (!image_sig_init(ctx, &data, image))
return 0;
image_sig_preprocess(&data);
if (data.valid)
{
image_sig_segmentation(&data);
image_sig_detect_textured(&data);
}
image_sig_finish(&data, sig);
image_sig_cleanup(&data);
return 1;
}