def pSLID(img, thresh=150):
"""find all lines using different settings"""
print(utils.call("pSLID(img)"))
segments = []; i = 0
for key, arr in enumerate(NC_SLID_CLAHE):
tmp = slid_clahe(img, limit=arr[0], grid=arr[1], iters=arr[2])
__segments = list(slid_detector(slid_canny(tmp), thresh))
segments += __segments; i += 1
print("FILTER: {} {} : {}".format(i, arr, len(__segments)))
debug.image(slid_canny(tmp)).lines(__segments).save("pslid_F%d" % i)
return segments
def slid_clahe(img, limit=2, grid=(3,3), iters=5):
"""repair using CLAHE algorithm (adaptive histogram equalization)"""
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
for i in range(iters):
img = cv2.createCLAHE(clipLimit=limit, \
tileGridSize=grid).apply(img)
debug.image(img).save("slid_clahe_@1")
if limit != 0:
kernel = np.ones((10, 10), np.uint8)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
debug.image(img).save("slid_clahe_@2")
return img
def llr_pad(four_points, img):
#print(utils.call("llr_pad(four_points)"))
pco = pyclipper.PyclipperOffset()
pco.AddPath(four_points, pyclipper.JT_MITER, pyclipper.ET_CLOSEDPOLYGON)
padded = pco.Execute(60)[0]
debug.image(img) \
.points(four_points, color=(0,0,255)) \
.points(padded, color=(0,255,0)) \
.lines([[four_points[0], four_points[1]], [four_points[1], four_points[2]], \
[four_points[2], four_points[3]], [four_points[3], four_points[0]]], \
color=(255,255,255)) \
.lines([[padded[0], padded[1]], [padded[1], padded[2]], \
[padded[2], padded[3]], [padded[3], padded[0]]], \
color=(255,255,255)) \
.save("llr_final_pad")
return pco.Execute(60)[0] # 60,70/75 is best (with buffer/for debug purpose)
def LAPS(img, lines, size=10):
#print(utils.call("LAPS(img, lines)"))
__points, points = laps_intersections(lines), []
debug.image(img).points(__points, size=3).save("laps_in_queue")
for pt in __points:
# pixels are in integers
pt = list(map(int, pt))
# size of our analysis area
lx1 = max(0, int(pt[0] - size - 1))
lx2 = max(0, int(pt[0] + size))
ly1 = max(0, int(pt[1] - size))
ly2 = max(0, int(pt[1] + size + 1))
# cropping for detector
dimg = img[ly1:ly2, lx1:lx2]
dimg_shape = np.shape(dimg)
# not valid
if dimg_shape[0] <= 0 or dimg_shape[1] <= 0: continue
# use neural network
re_laps = laps_detector(dimg)
if not re_laps[0]: continue
# add if okay
if pt[0] < 0 or pt[1] < 0: continue
points += [pt]
points = laps_cluster(points)
debug.image(img).points(points, size=5, \
color=debug.color()).save("laps_good_points")
return points
def SLID(img, segments):
# FIXME: zrobic 2 rodzaje haszowania (katy + pasy [blad - delta])
print(utils.call("SLID(img, segments)"))
global all_points; all_points = []
pregroup, group, hashmap, raw_lines = [[], []], {}, {}, []
__cache = {}
def __dis(a, b):
idx = hash("__dis" + str(a) + str(b))
if idx in __cache: return __cache[idx]
__cache[idx] = np.linalg.norm(na(a)-na(b))
return __cache[idx]
X = {}
def __fi(x):
if x not in X: X[x] = 0;
if (X[x] == x or X[x] == 0): X[x] = x
else: X[x] = __fi(X[x])
return X[x]
def __un(a, b):
ia, ib = __fi(a), __fi(b)
X[ia] = ib; group[ib] |= group[ia]
#group[ia] = set()
#group[ia] = set()
# shortest path // height
nln = lambda l1, x, dx: \
np.linalg.norm(np.cross(na(l1[1])-na(l1[0]),
na(l1[0])-na( x)))/dx
def __similar(l1, l2):
da, db = __dis(l1[0], l1[1]), __dis(l2[0], l2[1])
# if da > db: l1, l2, da, db = l2, l1, db, da
d1a, d2a = nln(l1, l2[0], da), nln(l1, l2[1], da)
d1b, d2b = nln(l2, l1[0], db), nln(l2, l1[1], db)
ds = 0.25 * (d1a + d1b + d2a + d2b) + 0.00001
#print(da, db, abs(da-db))
#print(int(da/ds), int(db/ds), "|", int(abs(da-db)), int(da+db),
# int(da+db)/(int(abs(da-db))+0.00001))
alfa = 0.0625 * (da + db) #15
# FIXME: roznica???
#if d1 + d2 == 0: d1 += 0.00001 # [FIXME]: divide by 0
t1 = (da/ds > alfa and db/ds > alfa)
if not t1: return False # [FIXME]: dist???
return True
def __generate(a, b, n):
points = []; t = 1/n
for i in range(n):
x = a[0] + (b[0]-a[0]) * (i * t)
y = a[1] + (b[1]-a[1]) * (i * t)
points += [[int(x), int(y)]]
return points
def __analyze(group):
global all_points
points = []
for idx in group:
points += __generate(*hashmap[idx], 10)
_, radius = cv2.minEnclosingCircle(na(points)); w = radius * (math.pi/2)
vx, vy, cx, cy = cv2.fitLine(na(points), cv2.DIST_L2, 0, 0.01, 0.01)
# debug.color()
all_points += points
return [[int(cx-vx*w), int(cy-vy*w)], [int(cx+vx*w), int(cy+vy*w)]]
for l in segments:
h = hash(str(l))
t1 = l[0][0] - l[1][0]
t2 = l[0][1] - l[1][1]
hashmap[h] = l; group[h] = set([h]); X[h] = h
if abs(t1) < abs(t2): pregroup[0].append(l)
else: pregroup[1].append(l)
debug.image(img.shape) \
.lines(pregroup[0], color=debug.color()) \
.lines(pregroup[1], color=debug.color()) \
.save("slid_pre_groups")
for lines in pregroup:
for i in range(len(lines)):
l1 = lines[i]; h1 = hash(str(l1))
#print(h1, __fi(h1))
if (X[h1] != h1): continue
#if (__fi(h1) != h1): continue
for j in range(i+1, len(lines)):
l2 = lines[j]; h2 = hash(str(l2))
#if (__fi(h2) != h2): continue
if (X[h2] != h2): continue
#if (len(group[h2])==0): continue
if not __similar(l1, l2): continue
__un(h1, h2) # union & find
# break # FIXME
__d = debug.image(img.shape)
for i in group:
#if (__fi(i) != i): continue
if (X[i] != i): continue
#if len(group[i]) == 0: continue
ls = [hashmap[h] for h in group[i]]
__d.lines(ls, color=debug.color())
__d.save("slid_all_groups")
for i in group:
#if (__fi(i) != i): continue
if (X[i] != i): continue
#if len(group[i]) == 0: continue
#if (__fi(i) != i): continue
raw_lines += [__analyze(group[i])]
debug.image(img.shape).lines(raw_lines).save("slid_final")
debug.image(img.shape)\
.points(all_points, color=(0,255,0), size=2)\
.lines(raw_lines).save("slid_final2")
return raw_lines
def LLR(img, points, lines):
#print(utils.call("LLR(img, points, lines)"))
old = points
# --- otoczka
def __convex_approx(points, alfa=0.01):
hull = scipy.spatial.ConvexHull(na(points)).vertices
cnt = na([points[pt] for pt in hull])
approx = cv2.approxPolyDP(cnt,alfa*\
cv2.arcLength(cnt,True),True)
return llr_normalize(itertools.chain(*approx))
# ---
# --- geometria
__cache = {}
def __dis(a, b):
idx = hash("__dis" + str(a) + str(b))
if idx in __cache: return __cache[idx]
__cache[idx] = np.linalg.norm(na(a)-na(b))
return __cache[idx]
nln = lambda l1, x, dx: \
np.linalg.norm(np.cross(na(l1[1])-na(l1[0]),
na(l1[0])-na( x)))/dx
# ---
pregroup = [[], []] # podzial na 2 grupy (dla ramki)
S = {} # ranking ramek // wraz z wynikiem
points = llr_correctness(llr_normalize(points), img.shape) # popraw punkty
# --- clustrowanie
__points = {}; points = llr_polysort(points); __max, __points_max = 0, []
alfa = math.sqrt(cv2.contourArea(na(points))/49)
X = sklearn.cluster.DBSCAN(eps=alfa*4).fit(points) # **(1.3)
for i in range(len(points)): __points[i] = []
for i in range(len(points)):
if X.labels_[i] != -1: __points[X.labels_[i]] += [points[i]]
for i in range(len(points)):
if len(__points[i]) > __max:
__max = len(__points[i]); __points_max = __points[i]
if len(__points) > 0 and len(points) > 49/2: points = __points_max
#print(X.labels_)
# ---
# tworzymy zewnetrzny pierscien
ring = __convex_approx(llr_polysort(points))
n = len(points); beta = n*(5/100) # beta=n*(100-(skutecznosc LAPS))
alfa = math.sqrt(cv2.contourArea(na(points))/49) # srednia otoczka siatki
x = [p[0] for p in points] # szukamy punktu
y = [p[1] for p in points] # centralnego skupiska
centroid = (sum(x) / len(points), \
sum(y) / len(points))
#print(alfa, beta, centroid)
# C (x2, y2) d=(x_1−x_0)^2+(y_1−y_0)^2, t=d_t/d
# B (x1, y1) (x_2,y_2)=(((1−t)x_0+tx_1),((1−t)y_0+ty_1))
# . t=(x_0-x_2)/(x_0-x_1)
# .
# A (x0, y0)
def __v(l):
y_0, x_0 = l[0][0], l[0][1]
y_1, x_1 = l[1][0], l[1][1]
x_2 = 0; t=(x_0-x_2)/(x_0-x_1+0.0001)
a = [int((1-t)*x_0+t*x_1), int((1-t)*y_0+t*y_1)][::-1]
x_2 = img.shape[0]; t=(x_0-x_2)/(x_0-x_1+0.0001)
b = [int((1-t)*x_0+t*x_1), int((1-t)*y_0+t*y_1)][::-1]
poly1 = llr_polysort([[0,0], [0, img.shape[0]], a, b])
s1 = llr_polyscore(na(poly1), points, centroid, beta=beta, alfa=alfa/2)
poly2 = llr_polysort([a, b, \
[img.shape[1],0], [img.shape[1],img.shape[0]]])
s2 = llr_polyscore(na(poly2), points, centroid, beta=beta, alfa=alfa/2)
return [a, b], s1, s2
def __h(l):
x_0, y_0 = l[0][0], l[0][1]
x_1, y_1 = l[1][0], l[1][1]
x_2 = 0; t=(x_0-x_2)/(x_0-x_1+0.0001)
a = [int((1-t)*x_0+t*x_1), int((1-t)*y_0+t*y_1)]
x_2 = img.shape[1]; t=(x_0-x_2)/(x_0-x_1+0.0001)
b = [int((1-t)*x_0+t*x_1), int((1-t)*y_0+t*y_1)]
poly1 = llr_polysort([[0,0], [img.shape[1], 0], a, b])
s1 = llr_polyscore(na(poly1), points, centroid, beta=beta, alfa=alfa/2)
poly2 = llr_polysort([a, b, \
[0, img.shape[0]], [img.shape[1], img.shape[0]]])
s2 = llr_polyscore(na(poly2), points, centroid, beta=beta, alfa=alfa/2)
return [a, b], s1, s2
for l in lines: # bedziemy wszystkie przegladac
for p in points: # odrzucamy linie ktore nie pasuja
# (1) linia przechodzi blisko dobrego punktu
t1 = nln(l, p, __dis(*l)) < alfa
# (2) linia przechodzi przez srodek skupiska
t2 = nln(l, centroid, __dis(*l)) > alfa * 2.5 # 3
# (3) linia nalezy do pierscienia
# t3 = True if p in ring else False
if t1 and t2:
#if (t1 and t2) or (t1 and t3 and t2): # [1 and 2] or [1 and 3 and 2]
tx, ty = l[0][0]-l[1][0], l[0][1]-l[1][1]
if abs(tx) < abs(ty): ll, s1, s2 = __v(l); o = 0
else: ll, s1, s2 = __h(l); o = 1
if s1 == 0 and s2 == 0: continue
pregroup[o] += [ll]
pregroup[0] = llr_unique(pregroup[0])
pregroup[1] = llr_unique(pregroup[1])
from laps import laps_intersections
debug.image(img) \
.lines(lines, color=(0,0,255)) \
.points(laps_intersections(lines), color=(255,0,0), size=2) \
.save("llr_debug_1")
debug.image(img) \
.points(laps_intersections(lines), color=(0,0,255), size=2) \
.points(old, color=(0,255,0)) \
.save("llr_debug_2")
debug.image(img) \
.lines(lines, color=(0,0,255)) \
.points(points, color=(0,0,255)) \
.points(ring, color=(0,255,0)) \
.points([centroid], color=(255,0,0)) \
.save("llr_debug")
debug.image(img) \
.lines(pregroup[0], color=(0,0,255)) \
.lines(pregroup[1], color=(255,0,0)) \
.save("llr_pregroups")
#print("---------------------")
for v in itertools.combinations(pregroup[0], 2): # poziome
for h in itertools.combinations(pregroup[1], 2): # pionowe
poly = laps_intersections([v[0], v[1], h[0], h[1]]) # przeciecia
poly = llr_correctness(poly, img.shape) # w obrazku
if len(poly) != 4: continue # jesl. nie ma
poly = na(llr_polysort(llr_normalize(poly))) # sortuj
if not cv2.isContourConvex(poly): continue # wypukly?
S[-llr_polyscore(poly, points, centroid, \
beta=beta, alfa=alfa/2)] = poly # dodaj
S = collections.OrderedDict(sorted(S.items())) # max
K = next(iter(S))
#print("key --", K)
four_points = llr_normalize(S[K]) # score
# XXX: pomijanie warst, lub ich wybor? (jesli mamy juz okay)
# XXX: wycinanie pod sam koniec? (modul wylicznia ile warstw potrzebnych)
#print("POINTS:", len(points))
#print("LINES:", len(lines))
debug.image(img).points(four_points).save("llr_four_points")
debug.image(img) \
.points(points, color=(0,255,0)) \
.points(four_points, color=(0,0,255)) \
.points([centroid], color=(255,0,0)) \
.lines([[four_points[0], four_points[1]], [four_points[1], four_points[2]], \
[four_points[2], four_points[3]], [four_points[3], four_points[0]]], \
color=(255,255,255)) \
.save("llr_debug_3")
return four_points
