class MLE:
def __init__(self, image, **kwargs):
self.connectivity = kwargs.pop("connectivity", "8")
self.filename = image
self.image = Image(filename=self.filename)
self.folder = kwargs.pop("folder", "chain")
self.guz, self.theta0 = self.loadChain()
self.gux = self.countChars()
self.theta = numpy.copy(self.theta0)
self.allThetas = {1: list(self.theta)}
def iteration(self):
num_u = numpy.array([0, 0, 0])
den_u = 0
for i in range(len(self.guz)):
num_u = num_u + self.guz[i]*numpy.exp(sum(self.guz[i]*self.theta - self.guz[i]*self.theta0))
den_u = den_u + numpy.exp(sum(self.guz[i]*self.theta - self.guz[i]*self.theta0))
u = self.gux - num_u/den_u
num_j = 0
den_j = 0
for i in range(len(self.guz)):
num_j = num_j + sum((self.guz[i] - num_u/den_u)**2)*numpy.exp(sum(self.guz[i]*self.theta - self.guz[i]*self.theta0))
den_j = den_j + numpy.exp(sum(self.guz[i]*self.theta - self.guz[i]*self.theta0))
j = num_j/den_j
self.theta = self.theta + u/j
self.allThetas[list(self.allThetas.keys())[-1]+1] = list(self.theta)
def loadChain(self):
with open(os.path.join(*self.folder.split(os.sep), "characteristics.json"), "r") as chars:
chain = json.loads(chars.read())
guz = list()
for i in range(len(chain.keys())-5):
chars = chain[str(i+1)]
guz.append(numpy.array([chars[0], chars[1], chars[2][0]-chars[2][1]]))
return guz, numpy.array(chain["params"])
def saveThetas(self):
path = self.filename.split(os.sep)[:-1] + ["estimation.json"]
with open(os.sep.join(path), "w") as thetaFile:
thetaFile.write(json.dumps(self.allThetas, sort_keys=True, indent=4))
def countChars(self):
a = self.image.area()
l = self.image.length()
chi = self.image.chi(self.connectivity)
return numpy.array([a, l, chi[0]-chi[1]])
class MH:
def __init__(self, **kwargs):
self.connectivity = kwargs.pop("connectivity", 8)
self.size = kwargs.pop("size", 100)
self.iters = kwargs.pop("iters", 1000)
self.r = kwargs.pop("r", (5, 10))
self.intensity = kwargs.pop("intensity", 1)
self.params = numpy.array(list(kwargs.pop("params", (0, 0, 0))))
self.image = Image(matrix=numpy.ones((self.size, self.size), dtype=numpy.uint8)*255)
self.circles = list()
def iteration(self):
action = "add" if random.random() < 0.5 or len(self.circles) == 0 else "remove"
if action == "add":
circle = self.drawCircle()
if random.random() < self.HR(action):
self.circles.append(circle)
self.commit()
else:
index = self.removeCircle()
if random.random() > self.HR(action):
del self.circles[index]
self.commit()
return self.image.matrix, [self.image.area(), self.image.length(), self.image.chi(self.connectivity)]
def makeImage(self, circles):
image = Image(matrix=numpy.ones((self.size, self.size), dtype=numpy.uint8)*255)
for circle in circles:
cv2.circle(image.matrix, *circle, 0, -1)
return image
def drawCircle(self):
self.temp = Image(matrix=numpy.copy(self.image.matrix))
circle = self.randomCircle()
cv2.circle(self.temp.matrix, *circle, 0, -1)
return circle
def removeCircle(self):
index = random.randrange(len(self.circles))
self.temp = self.makeImage(self.circles[:index] + self.circles[index+1:])
return index
def randomCircle(self):
r = random.randrange(*self.r)
circle = ((random.randrange(r + 1, self.size - r), random.randrange(r + 1, self.size - r)), r)
return circle
def chars(self):
chi = self.image.chi(self.connectivity)
chars = [self.image.area(), self.image.length(), chi[0] - chi[1]]
return numpy.array(chars)
def tempChars(self):
chi = self.temp.chi(self.connectivity)
chars = [self.temp.area(), self.temp.length(), chi[0] - chi[1]]
return numpy.array(chars)
def H(self, G, n):
ratio = self.intensity*numpy.exp(numpy.sum(self.params*G))/(n + 1)
return ratio
def HR(self, action):
if action == "add":
G = self.tempChars() - self.chars()
ratio = self.H(G, len(self.circles))
else:
G = self.chars() - self.tempChars()
ratio = 1/self.H(G, len(self.circles) - 1)
return min(ratio, 1)
def commit(self):
self.image.matrix = numpy.copy(self.temp.matrix)
