|
a |
|
b/4x/batch_della.py |
|
|
1 |
# Files |
|
|
2 |
import os |
|
|
3 |
import pickle |
|
|
4 |
import sys |
|
|
5 |
|
|
|
6 |
# Basic |
|
|
7 |
import numpy as np |
|
|
8 |
|
|
|
9 |
# Image Processing |
|
|
10 |
import skimage as ski |
|
|
11 |
from skimage import io, feature, morphology, filter, exposure, color, transform, measure |
|
|
12 |
import scipy.signal as sig |
|
|
13 |
from scipy.ndimage import maximum_filter, minimum_filter, binary_fill_holes |
|
|
14 |
|
|
|
15 |
# Stats |
|
|
16 |
import scipy.stats as st |
|
|
17 |
|
|
|
18 |
# Visualisation |
|
|
19 |
#import matplotlib |
|
|
20 |
#import matplotlib.pyplot as plt |
|
|
21 |
#import seaborn |
|
|
22 |
|
|
|
23 |
|
|
|
24 |
|
|
|
25 |
|
|
|
26 |
# Results : sheep ID, entropy, entropic variance, lacunarity, directionality |
|
|
27 |
# IDs |
|
|
28 |
name = [] |
|
|
29 |
imageid = [] |
|
|
30 |
|
|
|
31 |
# Entropy |
|
|
32 |
ent = [] |
|
|
33 |
entstd = [] |
|
|
34 |
|
|
|
35 |
# Lacunarity |
|
|
36 |
lac = [] |
|
|
37 |
nonlac = [] |
|
|
38 |
|
|
|
39 |
# Gabor filtering |
|
|
40 |
direc = [] |
|
|
41 |
scale = [] |
|
|
42 |
|
|
|
43 |
# Inflammatory foci count and size |
|
|
44 |
inflcount = [] |
|
|
45 |
inflsize = [] |
|
|
46 |
inflstd = [] |
|
|
47 |
|
|
|
48 |
# Tissue to sinusoid ratio |
|
|
49 |
ratio = [] |
|
|
50 |
|
|
|
51 |
|
|
|
52 |
|
|
|
53 |
|
|
|
54 |
|
|
|
55 |
|
|
|
56 |
|
|
|
57 |
|
|
|
58 |
# Image files to process |
|
|
59 |
|
|
|
60 |
# If we're running on all files |
|
|
61 |
if len(sys.argv) == 1 : |
|
|
62 |
|
|
|
63 |
# Read files |
|
|
64 |
files = [] |
|
|
65 |
directory = "data/" |
|
|
66 |
for i in os.listdir(directory) : |
|
|
67 |
if i.endswith(".jpg") : # if it's a jpg |
|
|
68 |
if not i.endswith("_processed.jpg") : # and isn't a processed image |
|
|
69 |
files.append(directory + i) # then add it to the list to be processed |
|
|
70 |
|
|
|
71 |
|
|
|
72 |
# Otherwise, we ran launcher.py and need to do select images only |
|
|
73 |
else : |
|
|
74 |
|
|
|
75 |
F = open("filelist%i.p" % int(sys.argv[1]), "r") |
|
|
76 |
files = pickle.load(F) |
|
|
77 |
F.close() |
|
|
78 |
|
|
|
79 |
|
|
|
80 |
|
|
|
81 |
|
|
|
82 |
|
|
|
83 |
|
|
|
84 |
|
|
|
85 |
|
|
|
86 |
|
|
|
87 |
|
|
|
88 |
|
|
|
89 |
# Iterate over files |
|
|
90 |
for f in files : |
|
|
91 |
|
|
|
92 |
print "Thread %s. Opening file : %s" % (sys.argv[1], f) |
|
|
93 |
|
|
|
94 |
|
|
|
95 |
# If it's not been processed before |
|
|
96 |
if not os.path.exists(f + "_processed.jpg") : |
|
|
97 |
|
|
|
98 |
### PROCESSING |
|
|
99 |
|
|
|
100 |
# Read the image |
|
|
101 |
A = io.imread(f) |
|
|
102 |
|
|
|
103 |
# Constrast enhancement |
|
|
104 |
print "Thread %s. Sigmoid transform for contrast." % sys.argv[1] |
|
|
105 |
B = exposure.adjust_sigmoid(A, gain=12) |
|
|
106 |
|
|
|
107 |
# Extract luminosity |
|
|
108 |
print "Thread %s. Generating luminosity." % sys.argv[1] |
|
|
109 |
C = color.rgb2xyz(B)[:, :, 1] |
|
|
110 |
|
|
|
111 |
# Apply adaptive thresholding |
|
|
112 |
print "Thread %s. Performing adaptive thresholding." % sys.argv[1] |
|
|
113 |
D = filter.threshold_adaptive(C, 301) |
|
|
114 |
|
|
|
115 |
# Clean |
|
|
116 |
print "Thread %s. Cleaning image." % sys.argv[1] |
|
|
117 |
E = morphology.remove_small_objects(~morphology.remove_small_objects(~D, 100), 100) |
|
|
118 |
|
|
|
119 |
# Save to disk |
|
|
120 |
io.imsave(f + "_processed.jpg", ski.img_as_float(E)) |
|
|
121 |
|
|
|
122 |
# Downsample for Gabor filtering |
|
|
123 |
print "Thread %s. Downsampling image." % sys.argv[1] |
|
|
124 |
Es = ski.img_as_float(transform.rescale(E, 0.25)) |
|
|
125 |
|
|
|
126 |
|
|
|
127 |
else : |
|
|
128 |
|
|
|
129 |
# Otherwise, we've processed it before, so read it in for speed |
|
|
130 |
A = io.imread(f) |
|
|
131 |
E = ski.img_as_float(io.imread(f + "_processed.jpg")) |
|
|
132 |
Es = ski.img_as_float(transform.rescale(E, 0.25)) |
|
|
133 |
print "Thread %s. Image already processed, downsampled." % sys.argv[1] |
|
|
134 |
|
|
|
135 |
|
|
|
136 |
## ID |
|
|
137 |
name.append(int(f.split("data/Sheep")[1].split("-")[0].split(".jpg")[0])) |
|
|
138 |
imageid.append(int(f.split("data/Sheep")[1].split("-")[2].split(".jpg")[0])) |
|
|
139 |
|
|
|
140 |
|
|
|
141 |
|
|
|
142 |
|
|
|
143 |
|
|
|
144 |
|
|
|
145 |
|
|
|
146 |
|
|
|
147 |
|
|
|
148 |
|
|
|
149 |
|
|
|
150 |
|
|
|
151 |
|
|
|
152 |
|
|
|
153 |
|
|
|
154 |
|
|
|
155 |
|
|
|
156 |
|
|
|
157 |
### GABOR FILTERING |
|
|
158 |
|
|
|
159 |
# Define scales and orientations to compute over |
|
|
160 |
pixelscales = np.arange(15, 55, 2) |
|
|
161 |
gaborscales = 4. / pixelscales # 2 to 20 pixels |
|
|
162 |
|
|
|
163 |
orientations = np.linspace(0, np.pi * 11./12., 12) # 0 to 180 degrees in 15 degree increments |
|
|
164 |
|
|
|
165 |
# Results array |
|
|
166 |
gabor = np.zeros((len(orientations), len(gaborscales))) |
|
|
167 |
|
|
|
168 |
# Perform Gabor filtering |
|
|
169 |
for i, iv in enumerate(orientations) : |
|
|
170 |
for j, jv in enumerate(gaborscales) : |
|
|
171 |
gaborReal, gaborImag = filter.gabor_filter(Es, jv, iv) |
|
|
172 |
gabor[i, j] = np.sqrt(np.sum(np.abs(gaborReal) ** 2) + np.sum(np.abs(gaborImag) ** 2)) # Return energy |
|
|
173 |
print "Thread %s. Gabor filtering. Completion : %f" % (sys.argv[1], (i / float(len(orientations)))) |
|
|
174 |
|
|
|
175 |
# Determine orientation-independent scale which fits best |
|
|
176 |
optimalscale = np.argmax(np.sum(gabor, axis = 0)) |
|
|
177 |
|
|
|
178 |
# At this scale, calculate directionality coefficient |
|
|
179 |
g = gabor[:, optimalscale] |
|
|
180 |
directionality = (g.max() - g.min()) / g.max() |
|
|
181 |
|
|
|
182 |
# Results |
|
|
183 |
scale.append(pixelscales[optimalscale]) |
|
|
184 |
direc.append(directionality) |
|
|
185 |
|
|
|
186 |
|
|
|
187 |
|
|
|
188 |
|
|
|
189 |
|
|
|
190 |
|
|
|
191 |
|
|
|
192 |
|
|
|
193 |
|
|
|
194 |
|
|
|
195 |
|
|
|
196 |
|
|
|
197 |
|
|
|
198 |
|
|
|
199 |
|
|
|
200 |
|
|
|
201 |
### LACUNARITY AND ENTROPY |
|
|
202 |
|
|
|
203 |
# Define scales over which to compute lacunarity and entropy |
|
|
204 |
|
|
|
205 |
scaleent = [] |
|
|
206 |
scaleentstd = [] |
|
|
207 |
#scalelac = [] |
|
|
208 |
roylac = [] |
|
|
209 |
|
|
|
210 |
|
|
|
211 |
# Characteristic scale |
|
|
212 |
s = pixelscales[optimalscale] |
|
|
213 |
|
|
|
214 |
# Generate a disk at this scale |
|
|
215 |
circle = morphology.disk(s) |
|
|
216 |
circlesize = circle.sum() |
|
|
217 |
|
|
|
218 |
# Convolve with image |
|
|
219 |
Y = sig.fftconvolve(E, circle, "valid") |
|
|
220 |
|
|
|
221 |
# Compute information entropy |
|
|
222 |
px = Y.ravel() / circlesize |
|
|
223 |
#px = np.reshape(Y[int(i) : int(E.shape[0]-i), int(i) : int(E.shape[1]-i)] / circlesize, (E.shape[0]-2*i)*(E.shape[1]-2*i),) |
|
|
224 |
py = 1. - px |
|
|
225 |
idx = np.logical_and(px > 1. / circlesize, px < 1.) |
|
|
226 |
entropy = - ( np.mean(px[idx] * np.log(px[idx])) + np.mean(py[idx] * np.log(py[idx])) ) |
|
|
227 |
entropystd = np.std(px[idx] * np.log(px[idx]) + py[idx] * np.log(py[idx])) |
|
|
228 |
|
|
|
229 |
# Compute normalised lacunarity |
|
|
230 |
lx = np.var(Y) / (np.mean(Y) ** 2) + 1 |
|
|
231 |
ly = np.var(1. - Y) / (np.mean(1. - Y) ** 2) + 1 |
|
|
232 |
# lacunarity = 2. - (1. / lx + 1. / ly) |
|
|
233 |
|
|
|
234 |
# Roy et al, J. Struct. Geol. 2010 |
|
|
235 |
|
|
|
236 |
# Results |
|
|
237 |
roylac.append((lx - 1.) / (1./np.mean(E) - 1.)) |
|
|
238 |
scaleent.append(entropy) |
|
|
239 |
scaleentstd.append(entropystd) |
|
|
240 |
# scalelac.append(lacunarity) |
|
|
241 |
|
|
|
242 |
print "Thread %s. Calculated entropy and lacunarity." % sys.argv[1] |
|
|
243 |
|
|
|
244 |
|
|
|
245 |
|
|
|
246 |
|
|
|
247 |
# Old code evaluates lacunarity and entropy at all scales |
|
|
248 |
# We only need it at the image's characteristic scale |
|
|
249 |
""" |
|
|
250 |
# Iterate over scales |
|
|
251 |
for i, s in enumerate(lacscales) : |
|
|
252 |
|
|
|
253 |
# Generate a disk at this scale |
|
|
254 |
circle = morphology.disk(s) |
|
|
255 |
circlesize = circle.sum() |
|
|
256 |
|
|
|
257 |
# Convolve with image |
|
|
258 |
Y = sig.fftconvolve(E, circle, "valid") |
|
|
259 |
|
|
|
260 |
# Compute information entropy |
|
|
261 |
px = Y.ravel() / circlesize |
|
|
262 |
#px = np.reshape(Y[int(i) : int(E.shape[0]-i), int(i) : int(E.shape[1]-i)] / circlesize, (E.shape[0]-2*i)*(E.shape[1]-2*i),) |
|
|
263 |
py = 1. - px |
|
|
264 |
idx = np.logical_and(px > 1. / circlesize, px < 1.) |
|
|
265 |
entropy = - ( np.mean(px[idx] * np.log(px[idx])) + np.mean(py[idx] * np.log(py[idx])) ) |
|
|
266 |
entropystd = np.std(px[idx] * np.log(px[idx]) + py[idx] * np.log(py[idx])) |
|
|
267 |
|
|
|
268 |
# Compute normalised lacunarity |
|
|
269 |
lx = np.var(Y) / (np.mean(Y) ** 2) + 1 |
|
|
270 |
ly = np.var(1. - Y) / (np.mean(1. - Y) ** 2) + 1 |
|
|
271 |
# lacunarity = 2. - (1. / lx + 1. / ly) |
|
|
272 |
|
|
|
273 |
# Roy et al, J. Struct. Geol. 2010 |
|
|
274 |
|
|
|
275 |
# Results |
|
|
276 |
roylac.append((lx - 1.) / (1./np.mean(E) - 1.)) |
|
|
277 |
scaleent.append(entropy) |
|
|
278 |
scaleentstd.append(entropystd) |
|
|
279 |
# scalelac.append(lacunarity) |
|
|
280 |
|
|
|
281 |
print "Thread %s. Calculating entropy and lacunarity. Completion : %f" % (sys.argv[1], (float(i) / len(lacscales))) |
|
|
282 |
""" |
|
|
283 |
|
|
|
284 |
nonlac.append(roylac) |
|
|
285 |
ent.append(scaleent) |
|
|
286 |
entstd.append(scaleentstd) |
|
|
287 |
# lac.append(scalelac) |
|
|
288 |
|
|
|
289 |
|
|
|
290 |
|
|
|
291 |
|
|
|
292 |
|
|
|
293 |
|
|
|
294 |
|
|
|
295 |
|
|
|
296 |
|
|
|
297 |
|
|
|
298 |
|
|
|
299 |
|
|
|
300 |
|
|
|
301 |
|
|
|
302 |
|
|
|
303 |
|
|
|
304 |
|
|
|
305 |
|
|
|
306 |
|
|
|
307 |
|
|
|
308 |
# Inflammatory focus count |
|
|
309 |
|
|
|
310 |
qstain = np.array([[.26451728, .5205347, .81183386], [.9199094, .29797825, .25489032], [.28947765, .80015373, .5253158]]) |
|
|
311 |
|
|
|
312 |
deconv = ski.img_as_float(color.separate_stains(transform.rescale(A, 0.25), np.linalg.inv(qstain))) |
|
|
313 |
|
|
|
314 |
|
|
|
315 |
|
|
|
316 |
subveins1 = \ |
|
|
317 |
morphology.remove_small_objects( |
|
|
318 |
filter.threshold_adaptive( |
|
|
319 |
filter.gaussian_filter( |
|
|
320 |
deconv[:, :, 2] / deconv[:, :, 0], |
|
|
321 |
11), |
|
|
322 |
250, offset = -0.13), |
|
|
323 |
60) |
|
|
324 |
|
|
|
325 |
subveins2 = \ |
|
|
326 |
morphology.remove_small_objects( |
|
|
327 |
filter.threshold_adaptive( |
|
|
328 |
filter.gaussian_filter( |
|
|
329 |
maximum_filter( |
|
|
330 |
deconv[:, :, 2] / deconv[:, :, 0], |
|
|
331 |
5), |
|
|
332 |
11), |
|
|
333 |
250, offset = -0.13), |
|
|
334 |
60) |
|
|
335 |
|
|
|
336 |
veins = \ |
|
|
337 |
maximum_filter( |
|
|
338 |
morphology.remove_small_objects( |
|
|
339 |
binary_fill_holes( |
|
|
340 |
morphology.binary_closing( |
|
|
341 |
np.logical_or(subveins1, subveins2), |
|
|
342 |
morphology.disk(25)), |
|
|
343 |
), |
|
|
344 |
250), |
|
|
345 |
27) |
|
|
346 |
|
|
|
347 |
|
|
|
348 |
|
|
|
349 |
|
|
|
350 |
|
|
|
351 |
|
|
|
352 |
rawinflammation = \ |
|
|
353 |
morphology.remove_small_objects( |
|
|
354 |
filter.threshold_adaptive( |
|
|
355 |
exposure.adjust_sigmoid( |
|
|
356 |
filter.gaussian_filter( |
|
|
357 |
exposure.equalize_adapthist( |
|
|
358 |
exposure.rescale_intensity( |
|
|
359 |
deconv[:, :, 1], |
|
|
360 |
out_range = (0, 1)), |
|
|
361 |
ntiles_y = 1), |
|
|
362 |
5), |
|
|
363 |
cutoff = 0.6), |
|
|
364 |
75, offset = -0.12), |
|
|
365 |
250) |
|
|
366 |
|
|
|
367 |
|
|
|
368 |
inflammation = \ |
|
|
369 |
maximum_filter( |
|
|
370 |
rawinflammation, |
|
|
371 |
29) |
|
|
372 |
|
|
|
373 |
|
|
|
374 |
|
|
|
375 |
print "Thread %s. Image segmentation complete." % sys.argv[1] |
|
|
376 |
|
|
|
377 |
|
|
|
378 |
|
|
|
379 |
|
|
|
380 |
total = veins + inflammation |
|
|
381 |
coloured = np.zeros_like(deconv) |
|
|
382 |
coloured[:, :, 1] = veins |
|
|
383 |
coloured[:, :, 2] = inflammation |
|
|
384 |
|
|
|
385 |
|
|
|
386 |
labelled, regions = measure.label(total, return_num = True) |
|
|
387 |
|
|
|
388 |
|
|
|
389 |
inflammationcount = 0 |
|
|
390 |
inflammationsize = [] |
|
|
391 |
|
|
|
392 |
|
|
|
393 |
|
|
|
394 |
for b in range(1, regions) : |
|
|
395 |
|
|
|
396 |
if (inflammation * (labelled == b)).sum() / ((veins * (labelled == b)).sum() + 1) > 0.5 : |
|
|
397 |
inflammationcount += 1 |
|
|
398 |
inflammationsize.append((rawinflammation * labelled == b).sum()) |
|
|
399 |
|
|
|
400 |
|
|
|
401 |
|
|
|
402 |
inflcount.append(inflammationcount) |
|
|
403 |
inflsize.append(np.mean(inflammationsize)) |
|
|
404 |
inflstd.append(np.std(inflammationsize)) |
|
|
405 |
|
|
|
406 |
|
|
|
407 |
|
|
|
408 |
|
|
|
409 |
|
|
|
410 |
|
|
|
411 |
|
|
|
412 |
|
|
|
413 |
|
|
|
414 |
|
|
|
415 |
# Tissue to sinusoid ratio |
|
|
416 |
|
|
|
417 |
tissue = np.sum(ski.img_as_bool(Es) * (1 - ski.img_as_bool(total))) |
|
|
418 |
sinusoids = np.sum(Es.shape[0] * Es.shape[1] - np.sum(ski.img_as_bool(total))) |
|
|
419 |
ratio.append(tissue.astype(float) / sinusoids) |
|
|
420 |
|
|
|
421 |
|
|
|
422 |
|
|
|
423 |
|
|
|
424 |
|
|
|
425 |
|
|
|
426 |
|
|
|
427 |
|
|
|
428 |
# Pickle results |
|
|
429 |
|
|
|
430 |
results = {} |
|
|
431 |
results["ratio"] = ratio |
|
|
432 |
results["inflcount"] = inflcount |
|
|
433 |
results["inflsize"] = inflsize |
|
|
434 |
results["inflstd"] = inflstd |
|
|
435 |
results["entropy"] = ent |
|
|
436 |
results["entstd"] = entstd |
|
|
437 |
results["directionality"] = direc |
|
|
438 |
results["scale"] = scale |
|
|
439 |
results["name"] = name |
|
|
440 |
results["ID"] = imageid |
|
|
441 |
results["roylac"] = nonlac |
|
|
442 |
|
|
|
443 |
F = open("results/thread_%s" % sys.argv[1], "w") |
|
|
444 |
pickle.dump(results, F) |
|
|
445 |
F.close() |
|
|
446 |
|
|
|
447 |
|
|
|
448 |
|
|
|
449 |
|
|
|
450 |
|
|
|
451 |
|
|
|
452 |
|
|
|
453 |
|
|
|
454 |
|
|
|
455 |
|
|
|
456 |
|
|
|
457 |
|
|
|
458 |
|
|
|
459 |
|
|
|
460 |
|