Commit 8bdc2a60fedf36c4fa80f5cf4f49dd0d827c5b62
1 parent
d0d68211
Exists in
master
and in
3 other branches
Adding mips.pyx to invesalius project
Showing
2 changed files
with
392 additions
and
0 deletions
Show diff stats
| ... | ... | @@ -0,0 +1,379 @@ |
| 1 | +#http://en.wikipedia.org/wiki/Local_maximum_intensity_projection | |
| 2 | +import numpy as np | |
| 3 | +cimport numpy as np | |
| 4 | +cimport cython | |
| 5 | + | |
| 6 | +from libc.math cimport floor, ceil, sqrt, fabs | |
| 7 | +from cython.parallel import prange | |
| 8 | + | |
| 9 | +DTYPE = np.uint8 | |
| 10 | +ctypedef np.uint8_t DTYPE_t | |
| 11 | + | |
| 12 | +DTYPE16 = np.int16 | |
| 13 | +ctypedef np.int16_t DTYPE16_t | |
| 14 | + | |
| 15 | +DTYPEF32 = np.float32 | |
| 16 | +ctypedef np.float32_t DTYPEF32_t | |
| 17 | + | |
| 18 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 19 | +def lmip(np.ndarray[DTYPE16_t, ndim=3] image, int axis, DTYPE16_t tmin, | |
| 20 | + DTYPE16_t tmax, np.ndarray[DTYPE16_t, ndim=2] out): | |
| 21 | + cdef DTYPE16_t max | |
| 22 | + cdef int start | |
| 23 | + cdef int sz = image.shape[0] | |
| 24 | + cdef int sy = image.shape[1] | |
| 25 | + cdef int sx = image.shape[2] | |
| 26 | + | |
| 27 | + # AXIAL | |
| 28 | + if axis == 0: | |
| 29 | + for x in xrange(sx): | |
| 30 | + for y in xrange(sy): | |
| 31 | + max = image[0, y, x] | |
| 32 | + if max >= tmin and max <= tmax: | |
| 33 | + start = 1 | |
| 34 | + else: | |
| 35 | + start = 0 | |
| 36 | + for z in xrange(sz): | |
| 37 | + if image[z, y, x] > max: | |
| 38 | + max = image[z, y, x] | |
| 39 | + | |
| 40 | + elif image[z, y, x] < max and start: | |
| 41 | + break | |
| 42 | + | |
| 43 | + if image[z, y, x] >= tmin and image[z, y, x] <= tmax: | |
| 44 | + start = 1 | |
| 45 | + | |
| 46 | + out[y, x] = max | |
| 47 | + | |
| 48 | + #CORONAL | |
| 49 | + elif axis == 1: | |
| 50 | + for z in xrange(sz): | |
| 51 | + for x in xrange(sx): | |
| 52 | + max = image[z, 0, x] | |
| 53 | + if max >= tmin and max <= tmax: | |
| 54 | + start = 1 | |
| 55 | + else: | |
| 56 | + start = 0 | |
| 57 | + for y in xrange(sy): | |
| 58 | + if image[z, y, x] > max: | |
| 59 | + max = image[z, y, x] | |
| 60 | + | |
| 61 | + elif image[z, y, x] < max and start: | |
| 62 | + break | |
| 63 | + | |
| 64 | + if image[z, y, x] >= tmin and image[z, y, x] <= tmax: | |
| 65 | + start = 1 | |
| 66 | + | |
| 67 | + out[z, x] = max | |
| 68 | + | |
| 69 | + #CORONAL | |
| 70 | + elif axis == 2: | |
| 71 | + for z in xrange(sz): | |
| 72 | + for y in xrange(sy): | |
| 73 | + max = image[z, y, 0] | |
| 74 | + if max >= tmin and max <= tmax: | |
| 75 | + start = 1 | |
| 76 | + else: | |
| 77 | + start = 0 | |
| 78 | + for x in xrange(sx): | |
| 79 | + if image[z, y, x] > max: | |
| 80 | + max = image[z, y, x] | |
| 81 | + | |
| 82 | + elif image[z, y, x] < max and start: | |
| 83 | + break | |
| 84 | + | |
| 85 | + if image[z, y, x] >= tmin and image[z, y, x] <= tmax: | |
| 86 | + start = 1 | |
| 87 | + | |
| 88 | + out[z, y] = max | |
| 89 | + | |
| 90 | + | |
| 91 | +cdef DTYPE16_t get_colour(DTYPE16_t vl, DTYPE16_t wl, DTYPE16_t ww): | |
| 92 | + cdef DTYPE16_t out_colour | |
| 93 | + cdef DTYPE16_t min_value = wl - (ww / 2) | |
| 94 | + cdef DTYPE16_t max_value = wl + (ww / 2) | |
| 95 | + if vl < min_value: | |
| 96 | + out_colour = min_value | |
| 97 | + elif vl > max_value: | |
| 98 | + out_colour = max_value | |
| 99 | + else: | |
| 100 | + out_colour = vl | |
| 101 | + | |
| 102 | + return out_colour | |
| 103 | + | |
| 104 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 105 | +@cython.cdivision(True) | |
| 106 | +cdef float get_opacity(DTYPE16_t vl, DTYPE16_t wl, DTYPE16_t ww) nogil: | |
| 107 | + cdef float out_opacity | |
| 108 | + cdef DTYPE16_t min_value = wl - (ww / 2) | |
| 109 | + cdef DTYPE16_t max_value = wl + (ww / 2) | |
| 110 | + if vl < min_value: | |
| 111 | + out_opacity = 0.0 | |
| 112 | + elif vl > max_value: | |
| 113 | + out_opacity = 1.0 | |
| 114 | + else: | |
| 115 | + out_opacity = 1.0/(max_value - min_value) * (vl - min_value) | |
| 116 | + | |
| 117 | + return out_opacity | |
| 118 | + | |
| 119 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 120 | +@cython.cdivision(True) | |
| 121 | +cdef float get_opacity_f32(DTYPEF32_t vl, DTYPE16_t wl, DTYPE16_t ww) nogil: | |
| 122 | + cdef float out_opacity | |
| 123 | + cdef DTYPE16_t min_value = wl - (ww / 2) | |
| 124 | + cdef DTYPE16_t max_value = wl + (ww / 2) | |
| 125 | + if vl < min_value: | |
| 126 | + out_opacity = 0.0 | |
| 127 | + elif vl > max_value: | |
| 128 | + out_opacity = 1.0 | |
| 129 | + else: | |
| 130 | + out_opacity = 1.0/(max_value - min_value) * (vl - min_value) | |
| 131 | + | |
| 132 | + return out_opacity | |
| 133 | + | |
| 134 | + | |
| 135 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 136 | +@cython.cdivision(True) | |
| 137 | +def mida(np.ndarray[DTYPE16_t, ndim=3] image, int axis, DTYPE16_t wl, | |
| 138 | + DTYPE16_t ww, np.ndarray[DTYPE16_t, ndim=2] out): | |
| 139 | + cdef int sz = image.shape[0] | |
| 140 | + cdef int sy = image.shape[1] | |
| 141 | + cdef int sx = image.shape[2] | |
| 142 | + | |
| 143 | + cdef DTYPE16_t min = image.min() | |
| 144 | + cdef DTYPE16_t max = image.max() | |
| 145 | + cdef DTYPE16_t vl | |
| 146 | + | |
| 147 | + cdef DTYPE16_t min_value = wl - (ww / 2) | |
| 148 | + cdef DTYPE16_t max_value = wl + (ww / 2) | |
| 149 | + | |
| 150 | + cdef float fmax=0.0 | |
| 151 | + cdef float fpi | |
| 152 | + cdef float dl | |
| 153 | + cdef float bt | |
| 154 | + | |
| 155 | + cdef float alpha | |
| 156 | + cdef float alpha_p = 0.0 | |
| 157 | + cdef float colour | |
| 158 | + cdef float colour_p = 0 | |
| 159 | + | |
| 160 | + cdef int x, y, z | |
| 161 | + | |
| 162 | + # AXIAL | |
| 163 | + if axis == 0: | |
| 164 | + for x in prange(sx, nogil=True): | |
| 165 | + for y in xrange(sy): | |
| 166 | + fmax = 0.0 | |
| 167 | + alpha_p = 0.0 | |
| 168 | + colour_p = 0.0 | |
| 169 | + for z in xrange(sz): | |
| 170 | + vl = image[z, y, x] | |
| 171 | + fpi = 1.0/(max - min) * (vl - min) | |
| 172 | + if fpi > fmax: | |
| 173 | + dl = fpi - fmax | |
| 174 | + fmax = fpi | |
| 175 | + else: | |
| 176 | + dl = 0.0 | |
| 177 | + | |
| 178 | + bt = 1.0 - dl | |
| 179 | + | |
| 180 | + colour = fpi | |
| 181 | + alpha = get_opacity(vl, wl, ww) | |
| 182 | + colour = (bt * colour_p) + (1 - bt * alpha_p) * colour * alpha | |
| 183 | + alpha = (bt * alpha_p) + (1 - bt * alpha_p) * alpha | |
| 184 | + | |
| 185 | + colour_p = colour | |
| 186 | + alpha_p = alpha | |
| 187 | + | |
| 188 | + if alpha >= 1.0: | |
| 189 | + break | |
| 190 | + | |
| 191 | + | |
| 192 | + #out[y, x] = <DTYPE16_t>((max_value - min_value) * colour + min_value) | |
| 193 | + out[y, x] = <DTYPE16_t>((max - min) * colour + min) | |
| 194 | + | |
| 195 | + | |
| 196 | + #CORONAL | |
| 197 | + elif axis == 1: | |
| 198 | + for z in prange(sz, nogil=True): | |
| 199 | + for x in xrange(sx): | |
| 200 | + fmax = 0.0 | |
| 201 | + alpha_p = 0.0 | |
| 202 | + colour_p = 0.0 | |
| 203 | + for y in xrange(sy): | |
| 204 | + vl = image[z, y, x] | |
| 205 | + fpi = 1.0/(max - min) * (vl - min) | |
| 206 | + if fpi > fmax: | |
| 207 | + dl = fpi - fmax | |
| 208 | + fmax = fpi | |
| 209 | + else: | |
| 210 | + dl = 0.0 | |
| 211 | + | |
| 212 | + bt = 1.0 - dl | |
| 213 | + | |
| 214 | + colour = fpi | |
| 215 | + alpha = get_opacity(vl, wl, ww) | |
| 216 | + colour = (bt * colour_p) + (1 - bt * alpha_p) * colour * alpha | |
| 217 | + alpha = (bt * alpha_p) + (1 - bt * alpha_p) * alpha | |
| 218 | + | |
| 219 | + colour_p = colour | |
| 220 | + alpha_p = alpha | |
| 221 | + | |
| 222 | + if alpha >= 1.0: | |
| 223 | + break | |
| 224 | + | |
| 225 | + out[z, x] = <DTYPE16_t>((max - min) * colour + min) | |
| 226 | + | |
| 227 | + #AXIAL | |
| 228 | + elif axis == 2: | |
| 229 | + for z in prange(sz, nogil=True): | |
| 230 | + for y in xrange(sy): | |
| 231 | + fmax = 0.0 | |
| 232 | + alpha_p = 0.0 | |
| 233 | + colour_p = 0.0 | |
| 234 | + for x in xrange(sx): | |
| 235 | + vl = image[z, y, x] | |
| 236 | + fpi = 1.0/(max - min) * (vl - min) | |
| 237 | + if fpi > fmax: | |
| 238 | + dl = fpi - fmax | |
| 239 | + fmax = fpi | |
| 240 | + else: | |
| 241 | + dl = 0.0 | |
| 242 | + | |
| 243 | + bt = 1.0 - dl | |
| 244 | + | |
| 245 | + colour = fpi | |
| 246 | + alpha = get_opacity(vl, wl, ww) | |
| 247 | + colour = (bt * colour_p) + (1 - bt * alpha_p) * colour * alpha | |
| 248 | + alpha = (bt * alpha_p) + (1 - bt * alpha_p) * alpha | |
| 249 | + | |
| 250 | + colour_p = colour | |
| 251 | + alpha_p = alpha | |
| 252 | + | |
| 253 | + if alpha >= 1.0: | |
| 254 | + break | |
| 255 | + | |
| 256 | + out[z, y] = <DTYPE16_t>((max - min) * colour + min) | |
| 257 | + | |
| 258 | + | |
| 259 | + | |
| 260 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 261 | +@cython.cdivision(True) | |
| 262 | +cdef inline void finite_difference(DTYPE16_t[:, :, :] image, | |
| 263 | + int x, int y, int z, float h, float *g) nogil: | |
| 264 | + cdef int px, py, pz, fx, fy, fz | |
| 265 | + | |
| 266 | + cdef int sz = image.shape[0] | |
| 267 | + cdef int sy = image.shape[1] | |
| 268 | + cdef int sx = image.shape[2] | |
| 269 | + | |
| 270 | + cdef float gx, gy, gz | |
| 271 | + | |
| 272 | + if x == 0: | |
| 273 | + px = 0 | |
| 274 | + fx = 1 | |
| 275 | + elif x == sx - 1: | |
| 276 | + px = x - 1 | |
| 277 | + fx = x | |
| 278 | + else: | |
| 279 | + px = x - 1 | |
| 280 | + fx = x + 1 | |
| 281 | + | |
| 282 | + if y == 0: | |
| 283 | + py = 0 | |
| 284 | + fy = 1 | |
| 285 | + elif y == sy - 1: | |
| 286 | + py = y - 1 | |
| 287 | + fy = y | |
| 288 | + else: | |
| 289 | + py = y - 1 | |
| 290 | + fy = y + 1 | |
| 291 | + | |
| 292 | + if z == 0: | |
| 293 | + pz = 0 | |
| 294 | + fz = 1 | |
| 295 | + elif z == sz - 1: | |
| 296 | + pz = z - 1 | |
| 297 | + fz = z | |
| 298 | + else: | |
| 299 | + pz = z - 1 | |
| 300 | + fz = z + 1 | |
| 301 | + | |
| 302 | + gx = (image[z, y, fx] - image[z, y, px]) / (2*h) | |
| 303 | + gy = (image[z, fy, x] - image[z, py, x]) / (2*h) | |
| 304 | + gz = (image[fz, y, x] - image[pz, y, x]) / (2*h) | |
| 305 | + | |
| 306 | + g[0] = gx | |
| 307 | + g[1] = gy | |
| 308 | + g[2] = gz | |
| 309 | + | |
| 310 | + | |
| 311 | + | |
| 312 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 313 | +@cython.cdivision(True) | |
| 314 | +cdef inline float calc_fcm_itensity(DTYPE16_t[:, :, :] image, | |
| 315 | + int x, int y, int z, float n, float* dir) nogil: | |
| 316 | + cdef float g[3] | |
| 317 | + finite_difference(image, x, y, z, 1.0, g) | |
| 318 | + cdef float gm = sqrt(g[0]*g[0] + g[1]*g[1] + g[2]*g[2]) | |
| 319 | + cdef float d = g[0]*dir[0] + g[1]*dir[1] + g[2]*dir[2] | |
| 320 | + cdef float sf = (1.0 - fabs(d/gm))**n | |
| 321 | + #alpha = get_opacity_f32(gm, wl, ww) | |
| 322 | + cdef float vl = gm * sf | |
| 323 | + return vl | |
| 324 | + | |
| 325 | +@cython.boundscheck(False) # turn of bounds-checking for entire function | |
| 326 | +@cython.cdivision(True) | |
| 327 | +def fast_countour_mip(np.ndarray[DTYPE16_t, ndim=3] image, | |
| 328 | + float n, | |
| 329 | + int axis, | |
| 330 | + DTYPE16_t wl, DTYPE16_t ww, | |
| 331 | + int tmip, | |
| 332 | + np.ndarray[DTYPE16_t, ndim=2] out): | |
| 333 | + cdef int sz = image.shape[0] | |
| 334 | + cdef int sy = image.shape[1] | |
| 335 | + cdef int sx = image.shape[2] | |
| 336 | + cdef float gm | |
| 337 | + cdef float alpha | |
| 338 | + cdef float sf | |
| 339 | + cdef float d | |
| 340 | + | |
| 341 | + cdef float* g | |
| 342 | + cdef float* dir = [ 0, 0, 0 ] | |
| 343 | + | |
| 344 | + cdef DTYPE16_t[:, :, :] vimage = image | |
| 345 | + cdef np.ndarray[DTYPE16_t, ndim=3] tmp = np.empty_like(image) | |
| 346 | + | |
| 347 | + cdef DTYPE16_t min = image.min() | |
| 348 | + cdef DTYPE16_t max = image.max() | |
| 349 | + cdef float fmin = <float>min | |
| 350 | + cdef float fmax = <float>max | |
| 351 | + cdef float vl | |
| 352 | + cdef DTYPE16_t V | |
| 353 | + | |
| 354 | + cdef int x, y, z | |
| 355 | + | |
| 356 | + if axis == 0: | |
| 357 | + dir[2] = 1.0 | |
| 358 | + elif axis == 1: | |
| 359 | + dir[1] = 1.0 | |
| 360 | + elif axis == 2: | |
| 361 | + dir[0] = 1.0 | |
| 362 | + | |
| 363 | + for z in prange(sz, nogil=True): | |
| 364 | + for y in range(sy): | |
| 365 | + for x in range(sx): | |
| 366 | + vl = calc_fcm_itensity(vimage, x, y, z, n, dir) | |
| 367 | + tmp[z, y, x] = <DTYPE16_t>vl | |
| 368 | + | |
| 369 | + cdef DTYPE16_t tmin = tmp.min() | |
| 370 | + cdef DTYPE16_t tmax = tmp.max() | |
| 371 | + | |
| 372 | + #tmp = ((max - min)/<float>(tmax - tmin)) * (tmp - tmin) + min | |
| 373 | + | |
| 374 | + if tmip == 0: | |
| 375 | + out[:] = tmp.max(axis) | |
| 376 | + elif tmip == 1: | |
| 377 | + lmip(tmp, axis, 700, 3033, out) | |
| 378 | + elif tmip == 2: | |
| 379 | + mida(tmp, axis, wl, ww, out) | ... | ... |
| ... | ... | @@ -0,0 +1,13 @@ |
| 1 | +from distutils.core import setup | |
| 2 | +from distutils.extension import Extension | |
| 3 | +from Cython.Distutils import build_ext | |
| 4 | + | |
| 5 | +import numpy | |
| 6 | + | |
| 7 | +setup( | |
| 8 | + cmdclass = {'build_ext': build_ext}, | |
| 9 | + ext_modules = [ Extension("invesalius/data/mips", ["invesalius/data/mips.pyx"], | |
| 10 | + include_dirs = [numpy.get_include()], | |
| 11 | + extra_compile_args=['-fopenmp'], | |
| 12 | + extra_link_args=['-fopenmp'],)] | |
| 13 | + ) | ... | ... |