[Python] Connect neighbouring curves that are less than 5 pixels of each other

I have an image like the following:

This is an image with pixel width of just 1 pixel. At places where the curve can go continuously either vertically or horizontally, I need the curves to be joined at those certain places, so that they seem as one long curve (I am writing the points traced by the curve to a CSV). Basically I want to divide the curves vertically and horizontally. So overlapping is not an issue even if it happens. I just want long curves in the end.

I tried a method that uses two for loops to do so, which kind of works, but takes a long time for processing because of the number of curves in this image.

sortedCoordinates = sorted(cornerPoints, key=lambda x: x[0][0])

jointCount = 0
duplicateListPath1, duplicateListPath2 = [], []
mainDetails = []
for item1 in sortedCoordinates:
for item2 in sortedCoordinates:
if item1 == item2:
continue
point1, point2 = item1[1], item2[0]
dist = euclideanDistance(point1, point2)
if dist <= 3:
item1, item2 = sorted(item1, key=lambda x:x[0]), sorted(item2, key=lambda x:x[0])
if item1 == item2:
continue
jointCount += 1
path1 = traceCurveJoin(copyTopHatCopy, item1[0][::-1], item1[1][::-1])
path2 = traceCurveJoin(copyTopHatCopy, item2[0][::-1], item2[1][::-1])

item1AngleSum, item2AngleSum = 0, 0
for i in range(len(path1) - 1):
item1AngleSum += math.degrees(math.atan2(path1[i + 1][1] - path1[1], path1[i + 1][0] - path1[0]))
item1Angle = item1AngleSum / len(path1)
for i in range(len(path2) - 1):
item2AngleSum += math.degrees(math.atan2(path2[i + 1][1] - path2[1], path2[i + 1][0] - path2[0]))
item2Angle = item2AngleSum / len(path2)

if item1Angle < 0:
item1Angle += 180
if item2Angle < 0:
item2Angle += 180
if abs(item1Angle - item2Angle) < 90:
if path1 in duplicateListPath1 or path2 in duplicateListPath2 or path1 in duplicateListPath1 or path2 in duplicateListPath2:
continue
if path1 in duplicateListPath2 or path2 in duplicateListPath1 or path1 in duplicateListPath2 or path2 in duplicateListPath1:
continue
duplicateListPath1.append(path1)
duplicateListPath2.append(path2)
mainDetails.append((path1, path2))
imageHeight, imageWidth = copyTopHatCopy.shape[:2]
imageAnother = np.zeros((imageHeight, imageWidth), np.uint8)
oneCount = 0
cutList = []

for i, (ePath1, ePath2) in enumerate(mainDetails):
if oneCount == 1:
try:
csvList.pop(len(csvList) - 1)
except:
pass
oneCount = 0
# with lockOCC:
# overallContourCount.value += 1
# finalContourCount = overallContourCount.value

path1Start, path1End = ePath1[0], ePath1[-1]
path2Start, path2End = ePath2[0], ePath2[-1]

distances = {
(path1Start, path2Start): euclideanDistance(path1Start, path2Start),
(path1Start, path2End): euclideanDistance(path1Start, path2End),
(path1End, path2Start): euclideanDistance(path1End, path2Start),
(path1End, path2End): euclideanDistance(path1End, path2End)
}

closestPair = min(distances, key=distances.get)
path1Reversed, path2Reversed = ePath1[::-1], ePath2[::-1]

if closestPair[0] == path1Start:
ePath1 = ePath1[::-1]
if closestPair[1] == path2End:
ePath2 = ePath2[::-1]

intermeadiatePoints = interpolatePoints(closestPair[0], closestPair[1])
joinedPath = np.vstack([ePath1, intermeadiatePoints, ePath2])


Is there any way to do it faster than what I have done?

Continue reading...