frg = frags[index2Choose]
relation = functionMisc.relatePoints(
adjusted2[len(adjusted2) // 3], frg[0])
transdEdge = functionMisc.rotateEdgeToEdge(
adjusted2[len(adjusted2) // 3],
adjusted2[math.floor((len(adjusted2) // 3)) + math.floor(sz)],
frg)
transdEdge2 = functionMisc.ratioTransfer(
adjusted2[math.floor((len(adjusted2) // 3)) + math.floor(sz)],
transdEdge)
if len(transdEdge2) == 0:
fails = fails + 1
continue
crossing = functionMisc.ecross(
outerContour, transdEdge2) or (functionMisc.ecross(
adjusted2[0:math.floor((len(adjusted2) // 3)) - 1:1],
transdEdge2) or functionMisc.ecross(
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz) + 1:len(adjusted2):1],
transdEdge2))
if not crossing:
adjusted2 = np.concatenate(
(np.concatenate(
(adjusted2[0:math.floor((len(adjusted2) // 3)):1],
transdEdge2)),
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz):len(adjusted2):1]))
rtdd = functionMisc.createDescriptorSet(
adjusted2, height, width)
def petsketch(img):
height, width, channels = img.shape
#img = cv2.fastNlMeansDenoisingColored(img,None,20,20,7,21)
img = cv2.fastNlMeansDenoisingColored(img, None, 10, 10, 7, 21)
edges = cv2.Canny(img, 70, 200)
kernel = np.ones((2, 2), np.uint8)
dilation = cv2.dilate(edges, kernel, iterations=1)
im3, Ocontours, hierarchy = cv2.findContours(dilation, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
c2 = []
forgotten = []
for cont in Ocontours:
#print(len(cont))
if len(cont) > 10:
c2.append(cont)
elif len(cont) > 4:
forgotten.append(cont[:len(cont) // 2])
contours = c2
#print(forgotten)
# top = []
# bot = []
# left = []
# right = []
# for x in range(0,width):
# top.append([x,0])
# bot.append([width-x,height])
# for y in range(0,height):
# right.append([width,y])
# left.append([0,height-y])
# outerContour = top + right + bot+ left
blank_image = np.zeros((height, width, 3), np.uint8)
# im = cv2.imread("ex3.png")
# #im = cv2.imread("circle2.png")
# height, width, channels = im.shape
# imgray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
# ret,thresh = cv2.threshold(imgray,127,255,0)
# height, width, channels = im.shape
# im3, contours, hierarchy = cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)
# blank_image = np.zeros((height,width,3), np.uint8)
# # startlistOfPoints = []
# # for ind, x in enumerate(contours[1]):
# # xval = x[0][0]
# # yval = x[0][1]
# # startlistOfPoints.append([yval, xval])
# # listOfPoints= np.array(startlistOfPoints)
# #blank_image = np.zeros(((height*4)//2,((width*4)//2),3), np.uint8)
outerContour = []
for ind, x in enumerate(contours[0]):
xval = x[0][0]
yval = x[0][1]
outerContour.append([yval, xval])
# for x in outerContour:
# xval = x[1]
# yval = x[0]
# blank_image[yval][xval]=[0,255,255]
folderName = "flatfish"
frags = testFrags
for cont in range(1, len(contours)):
# print("onto ",cont," out of ",len(contours),len(contours[cont]))
# sys.stdout.flush()
startlistOfPoints = []
for ind, x in enumerate(contours[cont]):
xval = x[0][0]
yval = x[0][1]
startlistOfPoints.append([yval, xval])
listOfPoints = np.array(startlistOfPoints)
steps = 10
#maxAttempts = 1
adjusted2 = listOfPoints
fails = 0
for x in range(0, steps):
# selection = []
# distances = []
crossing = True
while crossing:
if fails > 0:
fails = 0
break
#historic = adjusted2
if (len(adjusted2) < 10):
break
elif (len(adjusted2) < 100):
select = getRandomSection(len(adjusted2),
len(adjusted2) // 3,
len(adjusted2) // 6)
else:
select = getRandomSection(len(adjusted2),
len(adjusted2) // 10,
len(adjusted2) // 15)
adjusted2 = adjustRoundList(adjusted2, math.floor(select[1]),
math.floor(select[0]))
sz = 1
if (select[0] > select[1]):
sz = select[0] - select[1]
else:
sz = len(adjusted2) - select[1] + select[0]
subfrag = adjusted2[math.floor((
len(adjusted2) // 3)):math.floor((len(adjusted2) // 3)) +
math.floor(sz):1]
subdesc = generateDescriptor(subfrag, height, width)
myResult, myDistance = flann.nn_index(subdesc,
5,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
frg = frags[myResult[0][0]]
relation = functionMisc.relatePoints(
adjusted2[len(adjusted2) // 3], frg[0])
transdEdge = functionMisc.rotateEdgeToEdge(
adjusted2[len(adjusted2) // 3],
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz)], frg)
transdEdge2 = functionMisc.ratioTransfer(
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz)], transdEdge)
if len(transdEdge2) == 0:
fails = fails + 1
#print("lenfail length of frg ",len(frg), " sz ",sz)
#break
if fails < 1:
#crossing = functionMisc.ecross(outerContour,transdEdge2) or (functionMisc.ecross(adjusted2[0 : math.floor((len(adjusted2)//3))-1 : 1],transdEdge2) or functionMisc.ecross(adjusted2[math.floor((len(adjusted2)//3))+math.floor(sz)+1 : len(adjusted2) : 1], transdEdge2))
#CAUSE WE NO LONGER CARE
crossing = functionMisc.ecross(outerContour, transdEdge2)
if not crossing:
#transdEdge2
#adjusted2[0 : math.floor((len(adjusted2)//3)) : 1]
partReplacing = adjusted2[math.floor((
len(adjusted2) //
3)):math.floor((len(adjusted2) // 3)) +
math.floor(sz):1]
newTransdEdge = []
# maxp = 5
# petx = random.randint(-maxp,maxp) * random.randint(-maxp,maxp)
# pety = random.randint(-maxp,maxp) * random.randint(-maxp,maxp)
petx = 0
pety = 0
for ind in range(0, len(transdEdge2)):
p = transdEdge2[ind]
q = partReplacing[(len(partReplacing) * ind) //
len(transdEdge2)]
newTransdEdge.append([
petx + q[0] - (-p[0] + q[0]) // 1,
pety + q[1] - (-p[1] + q[1]) // 1
])
transdEdge2 = newTransdEdge
adjusted2 = np.concatenate(
(np.concatenate(
(adjusted2[0:math.floor((len(adjusted2) //
3)):1], transdEdge2)),
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz):len(adjusted2):1]))
for x in adjusted2:
xval = x[1]
yval = x[0]
blank_image[yval][xval] = [255, 255, 255]
#print("fails: ",fails)
for line in forgotten:
for x in line:
yval = x[0][1]
xval = x[0][0]
blank_image[yval][xval] = [255, 255, 255]
# cv2.imshow('Edges',blank_image)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
return blank_image
def complete():
truths = 0
folderName = "flatfish"
c = np.load("MPEG7dataset\\test\\extractions\\" + folderName +
"\\metadata.npy")
# just for dimensions
im = cv2.imread('MPEG7dataset\\test\\apple\\apple-1.png')
height, width, channels = im.shape
blank_image = np.zeros(((height * 5) // 2, ((width * 5) // 2), 3),
np.uint8)
#finding most relevant in cow using cat as vs
apples = np.load('MPEG7dataset\\test\\extractions\\apple\\Descriptors.npy')
bell = np.load('MPEG7dataset\\test\\extractions\\bell\\Descriptors.npy')
bone = np.load('MPEG7dataset\\test\\extractions\\Bone\\Descriptors.npy')
car = np.load('MPEG7dataset\\test\\extractions\\car\\Descriptors.npy')
carriage = np.load(
'MPEG7dataset\\test\\extractions\\carriage\\Descriptors.npy')
cellphone = np.load(
'MPEG7dataset\\test\\extractions\\cellular_phone\\Descriptors.npy')
child = np.load(
'MPEG7dataset\\test\\extractions\\children\\Descriptors.npy')
chopper = np.load(
'MPEG7dataset\\test\\extractions\\chopper\\Descriptors.npy')
face = np.load('MPEG7dataset\\test\\extractions\\face\\Descriptors.npy')
flatfish = np.load(
'MPEG7dataset\\test\\extractions\\flatfish\\Descriptors.npy')
fountain = np.load(
'MPEG7dataset\\test\\extractions\\fountain\\Descriptors.npy')
keys = np.load('MPEG7dataset\\test\\extractions\\key\\Descriptors.npy')
shoe = np.load('MPEG7dataset\\test\\extractions\\shoe\\Descriptors.npy')
watch = np.load('MPEG7dataset\\test\\extractions\\watch\\Descriptors.npy')
testImBig = np.concatenate([
apples, child, carriage, face, chopper, watch, bell, bone, car,
cellphone, fountain, keys, shoe, flatfish
],
axis=0)
chosen = flatfish
flann = FLANN()
flann.build_index(testImBig,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
#relavantest stores the fragment indicies with the highest relevance, and their indicies
relavantest = []
#looker is the bound on the length of relevantest
looker = 30
ImgNo = 0
for imager in c:
a = np.split(chosen, c[ImgNo])[1]
#num is the index of the descriptor des relative to the lowest index of the image
num = 0
for des in a:
#print(len(a))
myResult, myDistance = flann.nn_index(des,
50,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
highestNonCow = len(testImBig) - len(chosen)
inspectArray = []
for myR in range(len(myResult[0])):
test = myResult[0][myR] - highestNonCow
if test < 0:
inspectArray = np.split(myResult[0], [0, myR])[1]
break
else:
pass
numNearerfound = 0
if len(inspectArray) > 1:
dict = {'init': 'fake'}
for cand in inspectArray:
imgPlace = cand - (highestNonCow + 1)
imgID = 0
for imageBound in c:
if (imageBound[1] > imgPlace) and (imageBound[0] <
imgPlace):
imgID = imageBound[0]
break
if imgID in dict:
pass
else:
dict[imgID] = 1
numNearerfound = numNearerfound + 1
myLower = c[ImgNo][0]
tic = 0
inserted = False
#just an ordered assertion into relavantest
for pre in relavantest:
if tic > looker:
#if weve gone over the length of the thing
#just unsyrence fir the deletion part later
break
if pre[0] < numNearerfound:
#insertion at tic
relavantest.insert(tic, ([numNearerfound, num + myLower]))
inserted = True
if len(relavantest) > looker:
#to ensure we dont go over our specified length, we delete the last element
del relavantest[len(relavantest) - 1]
break
tic = tic + 1
if inserted == False and len(relavantest) < looker:
#initial filling of relecantest
relavantest.append(([numNearerfound, num + myLower]))
num = num + 1
ImgNo = ImgNo + 1
fragList = []
cowFrags = np.load('MPEG7dataset\\test\\extractions\\' + folderName +
'\\Fragments.npy')
for top in relavantest:
number = top[1]
fragList.append(cowFrags[number])
#phase 2===============================================================================================================================================================================
#=================================================================================================================
#=================================================================================================================
fullShape = AllRelGet()
im = cv2.imread("exampleImage.png")
height, width, channels = im.shape
flann.build_index(testImBig,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
imgray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 127, 255, 0)
height, width, channels = im.shape
im3, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
blank_image = np.zeros((height, width, 3), np.uint8)
c = np.load("MPEG7dataset\\test\\extractions\\" + folderName +
"\\metadata.npy")
outerContour = []
for ind, x in enumerate(contours[0]):
xval = x[0][0]
yval = x[0][1]
outerContour.append([yval, xval])
for x in outerContour:
xval = x[1]
yval = x[0]
blank_image[yval][xval] = [0, 255, 255]
frags = fragList
listOfPoints = fullShape
steps = 0
maxAttempts = 5
adjusted2 = listOfPoints
fails = 0
for x in range(0, steps):
selection = []
distances = []
print("\n")
for x in range(0, maxAttempts):
crossing = True
while crossing:
if fails > 8:
fails = 0
break
historic = adjusted2
select = getRandomSection(len(adjusted2),
len(adjusted2) // 12,
len(adjusted2) // 20)
adjusted2 = adjustRoundList(adjusted2, math.floor(select[1]),
math.floor(select[0]))
adjustedRot = adjusted2
#sz calculates from just select, could hapepn after the adjust round
sz = 1
if (select[0] > select[1]):
sz = select[0] - select[1]
else:
sz = len(adjusted2) - select[1] + select[0]
#choose a random fragment
index2Choose = math.floor((len(frags) * random.random()) // 1)
frg = frags[index2Choose]
relation = functionMisc.relatePoints(
adjusted2[len(adjusted2) // 3], frg[0])
transdEdge = functionMisc.rotateEdgeToEdge(
adjusted2[len(adjusted2) // 3],
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz)], frg)
transdEdge2 = functionMisc.ratioTransfer(
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz)], transdEdge)
if len(transdEdge2) == 0:
fails = fails + 1
continue
crossing = functionMisc.ecross(
outerContour, transdEdge2) or (functionMisc.ecross(
adjusted2[0:math.floor((len(adjusted2) // 3)) - 1:1],
transdEdge2) or functionMisc.ecross(
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz) + 1:len(adjusted2):1],
transdEdge2))
if not crossing:
adjusted2 = np.concatenate(
(np.concatenate(
(adjusted2[0:math.floor((len(adjusted2) // 3)):1],
transdEdge2)),
adjusted2[math.floor((len(adjusted2) // 3)) +
math.floor(sz):len(adjusted2):1]))
targetLength = len(frg)
rtdd = functionMisc.createDescriptorSet(
adjusted2, height, width, targetLength, 2.2)
totalSum = 0
for descr in rtdd:
#======= replaced with full descriptor
# myResult, myDistance = flann.nn_index(descr, 5, algorithm="kmeans", branching=32, iterations=7, checks=16)
# intersum = 0
# for indist in myDistance[0]:
# intersum = intersum + indist
# totalSum = totalSum + intersum
#============
#full relevance \/\/\/
#=============
myResult, myDistance = flann.nn_index(
descr,
50,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
highestNonCow = len(testImBig) - len(chosen)
#the place in our index of the last noncow, the remainder of the index is cows //^^^
inspectArray = []
for myR in range(len(myResult[0])):
test = myResult[0][myR] - highestNonCow
if test < 0:
inspectArray = np.split(myResult[0],
[0, myR])[1]
break
else:
pass
numNearerfound = 0
if len(inspectArray) > 1:
dict = {'init': 'fake'}
for cand in inspectArray:
imgPlace = cand - (highestNonCow + 1)
imgID = 0
for imageBound in c:
if (imageBound[1] > imgPlace) and (
imageBound[0] < imgPlace):
imgID = imageBound[0]
break
if imgID in dict:
pass
else:
dict[imgID] = 1
print("!")
numNearerfound = numNearerfound + 1
if numNearerfound == 0:
for myR in range(len(myResult[0])):
test = myResult[0][myR] - highestNonCow
if test > 0:
inspectArray = np.split(
myResult[0], [0, myR])[1]
break
else:
pass
if len(inspectArray) > 1:
dict = {'init': 'fake'}
for cand in inspectArray:
imgPlace = cand - (highestNonCow + 1)
imgID = 0
for imageBound in c:
if (imageBound[1] > imgPlace) and (
imageBound[0] < imgPlace):
imgID = imageBound[0]
break
if imgID in dict:
pass
else:
dict[imgID] = 1
#print("!")
numNearerfound = numNearerfound - 1
totalSum = totalSum - numNearerfound
#==============
selection.append(adjusted2)
rtdd = functionMisc.createDescriptorSet(
adjustedRot, height, width, targetLength, 2.2)
totalSumOld = 0
for descr in rtdd:
myResult, myDistance = flann.nn_index(
descr,
50,
algorithm="kmeans",
branching=32,
iterations=7,
checks=16)
highestNonCow = len(testImBig) - len(chosen)
inspectArray = []
for myR in range(len(myResult[0])):
test = myResult[0][myR] - highestNonCow
if test < 0:
inspectArray = np.split(myResult[0],
[0, myR])[1]
break
else:
pass
numNearerfound = 0
if len(inspectArray) > 1:
dict = {'init': 'fake'}
for cand in inspectArray:
imgPlace = cand - (highestNonCow + 1)
imgID = 0
for imageBound in c:
if (imageBound[1] > imgPlace) and (
imageBound[0] < imgPlace):
imgID = imageBound[0]
break
if imgID in dict:
pass
else:
dict[imgID] = 1
numNearerfound = numNearerfound + 1
if numNearerfound == 0:
for myR in range(len(myResult[0])):
test = myResult[0][myR] - highestNonCow
if test > 0:
inspectArray = np.split(
myResult[0], [0, myR])[1]
break
else:
pass
if len(inspectArray) > 1:
dict = {'init': 'fake'}
for cand in inspectArray:
imgPlace = cand - (highestNonCow + 1)
imgID = 0
for imageBound in c:
if (imageBound[1] > imgPlace) and (
imageBound[0] < imgPlace):
imgID = imageBound[0]
break
if imgID in dict:
pass
else:
dict[imgID] = 1
#print("!")
numNearerfound = numNearerfound - 1
totalSumOld = totalSumOld - numNearerfound
adjusted2 = historic
print('NewNearer:', -totalSum, " OldNearer:", -totalSumOld)
distances = distances + [totalSum - totalSumOld]
indexI = 0
best = distances[0]
for distInd in range(0, len(distances)):
if (distances[distInd] < best):
indexI = distInd
best = distances[distInd]
if (best < 0):
adjusted2 = selection[indexI]
print("tic")
print("fails: ", fails)
#================================================================
for x in fullShape:
xval = x[1]
yval = x[0]
blank_image[yval][xval] = [255, 0, 0]
for x in adjusted2:
xval = x[1]
yval = x[0]
blank_image[yval][xval] = [255, 255, 255]
cv2.imshow('Edges', blank_image)
print("ok finish")
print("truth", truths)
plt.minorticks_on()
plt.legend()
plt.xlabel('x')
plt.ylabel('y')
plt.show()
cv2.waitKey(0)
cv2.destroyAllWindows()
def addFragment(fragments, links, nextFrag):
indexToTest = 0
while (indexToTest < len(links)):
removedLink = links[indexToTest]
remShape = []
for x in range(indexToTest + 1, len(links)):
remShape = remShape + fragments[x] + links[x]
for x in range(0, indexToTest):
remShape = remShape + fragments[x] + links[x]
remShape = remShape + fragments[indexToTest]
works = (not functionMisc.ecross(remShape, nextFrag))
if not works:
indexToTest = indexToTest + 1
continue
a = remShape[0]
b = remShape[len(remShape) - 1]
c = nextFrag[0]
d = nextFrag[len(nextFrag) - 1]
extra1 = functionMisc.fill([c, b])
extra1 = extra1[1:len(extra1) - 1:1]
extra2 = functionMisc.fill([a, d])
extra2 = extra2[1:len(extra2) - 1:1]
fullShape = remShape + extra1[::-1] + nextFrag + extra2[::-1]
FirstFail = not ((not functionMisc.ecross(remShape, extra1)) and
(not functionMisc.ecross(extra1, extra2)) and
(not functionMisc.ecross(remShape, extra2)) and
(not functionMisc.ecross(nextFrag, extra1)) and
(not functionMisc.ecross(nextFrag, extra2)))
if (not FirstFail):
newFragments = fragments[:indexToTest + 1:1] + [
nextFrag
] + fragments[indexToTest + 1::1]
newLinks = links[:indexToTest:1] + [extra1[::-1]] + [
extra2[::-1]
] + links[indexToTest + 1::1]
return [True, newFragments, newLinks, fullShape]
extra1 = functionMisc.fill([c, a])
extra1 = extra1[1:len(extra1) - 1:1]
extra2 = functionMisc.fill([b, d])
extra2 = extra2[1:len(extra2) - 1:1]
fullShape = remShape + extra2 + nextFrag[::-1] + extra1
SecondFail = not ((not functionMisc.ecross(remShape, extra1)) and
(not functionMisc.ecross(extra1, extra2)) and
(not functionMisc.ecross(remShape, extra2)) and
(not functionMisc.ecross(nextFrag, extra1)) and
(not functionMisc.ecross(nextFrag, extra2)))
if (not SecondFail):
newFragments = fragments[:indexToTest + 1:1] + [
nextFrag[::-1]
] + fragments[indexToTest + 1::1]
newLinks = links[:indexToTest:1] + [extra2] + [
extra1
] + links[indexToTest + 1::1]
return [True, newFragments, newLinks, fullShape]
indexToTest = indexToTest + 1
return [False]
# #NOT USED ANYWAYdstr = descrips[index2Choose]
# frg = frags[index2Choose]
relation = functionMisc.relatePoints( adjusted2[len(adjusted2)//3],frg[0])
transdEdge = functionMisc.rotateEdgeToEdge(adjusted2[len(adjusted2)//3],adjusted2[math.floor((len(adjusted2)//3))+math.floor(sz)], frg)
transdEdge2 = functionMisc.ratioTransfer(adjusted2[math.floor((len(adjusted2)//3))+math.floor(sz)],transdEdge)
if len(transdEdge2) == 0:
fails = fails+1
print("WHYFAIL?????")
break
#crossing = functionMisc.ecross(outerContour,transdEdge2) or (functionMisc.ecross(adjusted2[0 : math.floor((len(adjusted2)//3))-1 : 1],transdEdge2) or functionMisc.ecross(adjusted2[math.floor((len(adjusted2)//3))+math.floor(sz)+1 : len(adjusted2) : 1], transdEdge2))
#CAUSE WE NO LONGER CARE
crossing = functionMisc.ecross(outerContour,transdEdge2)
if not crossing:
adjusted2 = np.concatenate((np.concatenate((adjusted2[0 : math.floor((len(adjusted2)//3)) : 1],transdEdge2)),adjusted2[math.floor((len(adjusted2)//3))+math.floor(sz) : len(adjusted2) : 1]))
# rtdd = functionMisc.createDescriptorSet(adjusted2,height,width)
# totalSum = 0
# for descr in rtdd:
# #======= replaced with full descriptor
# # myResult, myDistance = flann.nn_index(descr, 5, algorithm="kmeans", branching=32, iterations=7, checks=16)
# # intersum = 0
# # for indist in myDistance[0]:
# # intersum = intersum + indist
# # totalSum = totalSum + intersum
# #============
# #full relevance \/\/\/
# #=============