Graph Bi-partition - Normalized cuts - math

Can anyone explain me how to do cut in graph in the Normalized cuts algorithm descriped here: http://web.cs.ucdavis.edu/~bai/ECS231/returnsfinal/WangH.pdf (page 3 bottom)?
I have image, graph, solved eigenvalue problem, eigenvector of 2nd smallest eigenvalue. But I don't know how to cut graph.

You don't need to 'cut' the graph. All you need to do is split the current image in 2 sets, A and B with A = y_1(i) > t and B = y_1(i) <= t where y_1 refering to your algorithm is the second smallest eigenvector (also the solution for normalized cuts, stated in proposition 2) and t as the section 4.4 of the paper explains could be 0, mean, median, mode or the value of your choice, with different results as shown in section 5.
You only need the indexes of where that happens, as if they were 2 different binary masks for your original image. Then the ncut_partitions algorithm recursively calls itself with both A and B sets separately as the new images.
Follow "Algorithm 1 Recursive two-way cut" line by line.

Related

How to know 4 nodes (in one line) have same directions (or not)

If I have 4 nodes on a line, for example:
Main line is formed by nodes A(0,0) and B(5,0)
and there is a line on this main line which is formed by C(2,0) and D(4,0)
It is clear that vector AB has the same direction with vector CD, or in other words, vector BA has the same direction with vector DC.
I know it is simple, that in this case we just calculate delta x and then divided by each length.
Is there any common technique and in a "straight-way fashion"?
For example by comparing the gradients (which is in this case, it would be still valid, since both gradients are zero, but if the coordinates are conversed, the computation will be error, since the gradients cannot be computed by dividing by zero).
Thanks in advance.
If you want to know whether AB has the same orientation as CD, compute the dot product (B-A)*(D-C). It will be positive if they point in the same direction, negative if they point in opposite directions, and zero if one of the vectors is zero (or otherwise perpendicular to the other, but you assumed collinear points so that can't happen).

Finding the coordinates of points from distance matrix

I have a set of points (with unknow coordinates) and the distance matrix. I need to find the coordinates of these points in order to plot them and show the solution of my algorithm.
I can set one of these points in the coordinate (0,0) to simpify, and find the others. Can anyone tell me if it's possible to find the coordinates of the other points, and if yes, how?
Thanks in advance!
EDIT
Forgot to say that I need the coordinates on x-y only
The answers based on angles are cumbersome to implement and can't be easily generalized to data in higher dimensions. A better approach is that mentioned in my and WimC's answers here: given the distance matrix D(i, j), define
M(i, j) = 0.5*(D(1, j)^2 + D(i, 1)^2 - D(i, j)^2)
which should be a positive semi-definite matrix with rank equal to the minimal Euclidean dimension k in which the points can be embedded. The coordinates of the points can then be obtained from the k eigenvectors v(i) of M corresponding to non-zero eigenvalues q(i): place the vectors sqrt(q(i))*v(i) as columns in an n x k matrix X; then each row of X is a point. In other words, sqrt(q(i))*v(i) gives the ith component of all of the points.
The eigenvalues and eigenvectors of a matrix can be obtained easily in most programming languages (e.g., using GSL in C/C++, using the built-in function eig in Matlab, using Numpy in Python, etc.)
Note that this particular method always places the first point at the origin, but any rotation, reflection, or translation of the points will also satisfy the original distance matrix.
Step 1, arbitrarily assign one point P1 as (0,0).
Step 2, arbitrarily assign one point P2 along the positive x axis. (0, Dp1p2)
Step 3, find a point P3 such that
Dp1p2 ~= Dp1p3+Dp2p3
Dp1p3 ~= Dp1p2+Dp2p3
Dp2p3 ~= Dp1p3+Dp1p2
and set that point in the "positive" y domain (if it meets any of these criteria, the point should be placed on the P1P2 axis).
Use the cosine law to determine the distance:
cos (A) = (Dp1p2^2 + Dp1p3^2 - Dp2p3^2)/(2*Dp1p2* Dp1p3)
P3 = (Dp1p3 * cos (A), Dp1p3 * sin(A))
You have now successfully built an orthonormal space and placed three points in that space.
Step 4: To determine all the other points, repeat step 3, to give you a tentative y coordinate.
(Xn, Yn).
Compare the distance {(Xn, Yn), (X3, Y3)} to Dp3pn in your matrix. If it is identical, you have successfully identified the coordinate for point n. Otherwise, the point n is at (Xn, -Yn).
Note there is an alternative to step 4, but it is too much math for a Saturday afternoon
If for points p, q, and r you have pq, qr, and rp in your matrix, you have a triangle.
Wherever you have a triangle in your matrix you can compute one of two solutions for that triangle (independent of a euclidean transform of the triangle on the plane). That is, for each triangle you compute, it's mirror image is also a triangle that satisfies the distance constraints on p, q, and r. The fact that there are two solutions even for a triangle leads to the chirality problem: You have to choose the chirality (orientation) of each triangle, and not all choices may lead to a feasible solution to the problem.
Nevertheless, I have some suggestions. If the number entries is small, consider using simulated annealing. You could incorporate chirality into the annealing step. This will be slow for large systems, and it may not converge to a perfect solution, but for some problems it's the best you and do.
The second suggestion will not give you a perfect solution, but it will distribute the error: the method of least squares. In your case the objective function will be the error between the distances in your matrix, and actual distances between your points.
This is a math problem. To derive coordinate matrix X only given by its distance matrix.
However there is an efficient solution to this -- Multidimensional Scaling, that do some linear algebra. Simply put, it requires a pairwise Euclidean distance matrix D, and the output is the estimated coordinate Y (perhaps rotated), which is a proximation to X. For programming reason, just use SciKit.manifold.MDS in Python.
The "eigenvector" method given by the favourite replies above is very general and automatically outputs a set of coordinates as the OP requested, however I noticed that that algorithm does not even ask for a desired orientation (rotation angle) for the frame of the output points, the algorithm chooses that orientation all by itself!
People who use it might want to know at what angle the frame will be tipped before hand so I found an equation which gives the answer for the case of up to three input points, however I have not had time to generalize it to n-points and hope someone will do that and add it to this discussion. Here are the three angles the output sides will form with the x-axis as a function of the input side lengths:
angle side a = arcsin(sqrt(((c+b+a)*(c+b-a)*(c-b+a)*(-c+b+a)*(c^2-b^2)^2)/(a^4*((c^2+b^2-a^2)^2+(c^2-b^2)^2))))*180/Pi/2
angle side b = arcsin(sqrt(((c+b+a)*(c+b-a)*(c-b+a)*(-c+b+a)*(c^2+b^2-a^2)^2)/(4*b^4*((c^2+b^2-a^2)^2+(c^2-b^2)^2))))*180/Pi/2
angle side c = arcsin(sqrt(((c+b+a)*(c+b-a)*(c-b+a)*(-c+b+a)*(c^2+b^2-a^2)^2)/(4*c^4*((c^2+b^2-a^2)^2+(c^2-b^2)^2))))*180/Pi/2
Those equations also lead directly to a solution to the OP's problem of finding the coordinates for each point because: the side lengths are already given from the OP as the input, and my equations give the slope of each side versus the x-axis of the solution, thus revealing the vector for each side of the polygon answer, and summing those sides through vector addition up to a desired vertex will produce the coordinate of that vertex. So if anyone can extend my angle equations to handling beyond three input lengths (but I note: that might be impossible?), it might be a very fast way to the general solution of the OP's question, since slow parts of the algorithms that people gave above like "least square fitting" or "matrix equation solving" might be avoidable.

How do I calculate the "difference" between two sequences of points?

I have two sequences of length n and m. Each is a sequence of points of the form (x,y) and represent curves in an image. I need to find how different (or similar) these sequences are given that fact that
one sequence is likely longer than the other (i.e., one can be half or a quarter as long as the other, but if they trace approximately the same curve, they are the same)
these sequences could be in opposite directions (i.e., sequence 1 goes from left to right, while sequence 2 goes from right to left)
I looked into some difference estimates like Levenshtein as well as edit-distances in structural similarity matching for protein folding, but none of them seem to do the trick. I could write my own brute-force method but I want to know if there is a better way.
Thanks.
Do you mean that you are trying to match curves that have been translated in x,y coordinates? One technique from image processing is to use chain codes [I'm looking for a decent reference, but all I can find right now is this] to encode each sequence and then compare those chain codes. You could take the sum of the differences (modulo 8) and if the result is 0, the curves are identical. Since the sequences are of different lengths and don't necessarily start at the same relative location, you would have to shift one sequence and do this again and again, but you only have to create the chain codes once. The only way to detect if one of the sequences is reversed is to try both the forward and reverse of one of the sequences. If the curves aren't exactly alike, the sum will be greater than zero but it is not straightforward to tell how different the curves are simply from the sum.
This method will not be rotationally invariant. If you need a method that is rotationally invariant, you should look at Boundary-Centered Polar Encoding. I can't find a free reference for that, but if you need me to describe it, let me know.
A method along these lines might work:
For both sequences:
Fit a curve through the sequence. Make sure that you have a continuous one-to-one function from [0,1] to points on this curve. That is, for each (real) number between 0 and 1, this function returns a point on the curve belonging to it. By tracing the function for all numbers from 0 to 1, you get the entire curve.
One way to fit a curve would be to draw a straight line between each pair of consecutive points (it is not a nice curve, because it has sharp bends, but it might be fine for your purpose). In that case, the function can be obtained by calculating the total length of all the line segments (Pythagoras). The point on the curve corresponding to a number Y (between 0 and 1) corresponds to the point on the curve that has a distance Y * (total length of all line segments) from the first point on the sequence, measured by traveling over the line segments (!!).
Now, after we have obtained such a function F(double) for the first sequence, and G(double) for the second sequence, we can calculate the similarity as follows:
double epsilon = 0.01;
double curveDistanceSquared = 0.0;
for(double d=0.0;d<1.0;d=d+epsilon)
{
Point pointOnCurve1 = F(d);
Point pointOnCurve2 = G(d);
//alternatively, use G(1.0-d) to check whether the second sequence is reversed
double distanceOfPoints = pointOnCurve1.EuclideanDistance(pointOnCurve2);
curveDistanceSquared = curveDistanceSquared + distanceOfPoints * distanceOfPoints;
}
similarity = 1.0/ curveDistanceSquared;
Possible improvements:
-Find an improved way to fit the curves. Note that you still need the function that traces the curve for the above method to work.
-When calculating the distance, consider reparametrizing the function G in such a way that the distance is minimized. (This means you have an increasing function R, such that R(0) = 0 and R(1)=1,
but which is otherwise general. When calculating the distance you use
Point pointOnCurve1 = F(d);
Point pointOnCurve2 = G(R(d));
Subsequently, you try to choose R in such a way that the distance is minimized. (to see what happens, note that G(R(d)) also traces the curve)).
Why not do some sort of curve fitting procedure (least-squares whether it be ordinary or non-linear) and see if the coefficients on the shape parameters are the same. If you run it as a panel-data sort of model, there are explicit statistical tests whether sets of parameters are significantly different from one another. That would solve the problem of the the same curve but sampled at different resolutions.
Step 1: Canonicalize the orientation. For example, let's say that all curved start at the endpoint with lowest lexicographic order.
def inCanonicalOrientation(path):
return path if path[0]<path[-1] else reversed(path)
Step 2: You can either be roughly accurate, or very accurate. If you wish to be very accurate, calculate a spline, or fit both curves to a polynomial of appropriate degree, and compare coefficients. If you'd like just a rough estimate, do as follows:
def resample(path, numPoints)
pathLength = pathLength(path) #write this function
segments = generateSegments(path)
currentSegment = next(segments)
segmentsSoFar = [currentSegment]
for i in range(numPoints):
samplePosition = i/(numPoints-1)*pathLength
while samplePosition > pathLength(segmentsSoFar)+currentSegment.length:
currentSegment = next(segments)
segmentsSoFar.insert(currentSegment)
difference = samplePosition - pathLength(segmentsSoFar)
howFar = difference/currentSegment.length
yield Point((1-howFar)*currentSegment.start + (howFar)*currentSegment.end)
This can be modified from a linear resampling to something better.
def error(pathA, pathB):
pathA = inCanonicalOrientation(pathA)
pathB = inCanonicalOrientation(pathB)
higherResolution = max([len(pathA), len(pathB)])
resampledA = resample(pathA, higherResolution)
resampledB = resample(pathA, higherResolution)
error = sum(
abs(pointInA-pointInB)
for pointInA,pointInB in zip(pathA,pathB)
)
averageError = error / len(pathAorB)
normalizedError = error / Z(AorB)
return normalizedError
Where Z is something like the "diameter" of your path, perhaps the maximum Euclidean distance between any two points in a path.
I would use a curve-fitting procedure, but also throw in a constant term, i.e. 0 =B0 + B1*X + B2*Y + B3*X*Y + B4*X^2 etc. This would catch the translational variance and then you can do a statistical comparison of the estimated coefficients of the curves formed by the two sets of points as a way of classifying them. I'm assuming you'll have to do bi-variate interpolation if the data form arbitrary curves in the x-y plane.

3D Trilateration using given distances of unknown fixed points

I am new to this forum and not a native english speaker, so please be nice! :)
Here is the challenge I face at the moment:
I want to calculate the (approximate) relative coordinates of yet unknown points in a 3D euclidean space based on a set of given distances between 2 points.
In my first approach I want to ignore possible multiple solutions, just taking the first one by random.
e.g.:
given set of distances: (I think its creating a pyramid with a right-angled triangle as a base)
P1-P2-Distance
1-2-30
2-3-40
1-3-50
1-4-60
2-4-60
3-4-60
Step1:
Now, how do I calculate the relative coordinates for those points?
I figured that the first point goes to 0,0,0 so the second one is 30,0,0.
After that the third points can be calculated by finding the crossing of the 2 circles from points 1 and 2 with their distances to point 3 (50 and 40 respectively). How do I do that mathematically? (though I took these simple numbers for an easy representation of the situation in my mind). Besides I do not know how to get to the answer in a correct mathematical way the third point is at 30,40,0 (or 30,0,40 but i will ignore that).
But getting the fourth point is not as easy as that. I thought I have to use 3 spheres in calculate the crossing to get the point, but how do I do that?
Step2:
After I figured out how to calculate this "simple" example I want to use more unknown points... For each point there is minimum 1 given distance to another point to "link" it to the others. If the coords can not be calculated because of its degrees of freedom I want to ignore all possibilities except one I choose randomly, but with respect to the known distances.
Step3:
Now the final stage should be this: Each measured distance is a bit incorrect due to real life situation. So if there are more then 1 distances for a given pair of points the distances are averaged. But due to the imprecise distances there can be a difficulty when determining the exact (relative) location of a point. So I want to average the different possible locations to the "optimal" one.
Can you help me going through my challenge step by step?
You need to use trigonometry - specifically, the 'cosine rule'. This will give you the angles of the triangle, which lets you solve the 3rd and 4th points.
The rules states that
c^2 = a^2 + b^2 - 2abCosC
where a, b and c are the lengths of the sides, and C is the angle opposite side c.
In your case, we want the angle between 1-2 and 1-3 - the angle between the two lines crossing at (0,0,0). It's going to be 90 degrees because you have the 3-4-5 triangle, but let's prove:
50^2 = 30^2 + 40^2 - 2*30*40*CosC
CosC = 0
C = 90 degrees
This is the angle between the lines (0,0,0)-(30,0,0) and (0,0,0)- point 3; extend along that line the length of side 1-3 (which is 50) and you'll get your second point (0,50,0).
Finding your 4th point is slightly trickier. The most straightforward algorithm that I can think of is to firstly find the (x,y) component of the point, and from there the z component is straightforward using Pythagoras'.
Consider that there is a point on the (x,y,0) plane which sits directly 'below' your point 4 - call this point 5. You can now create 3 right-angled triangles 1-5-4, 2-5-4, and 3-5-4.
You know the lengths of 1-4, 2-4 and 3-4. Because these are right triangles, the ratio 1-4 : 2-4 : 3-4 is equal to 1-5 : 2-5 : 3-5. Find the point 5 using trigonometric methods - the 'sine rule' will give you the angles between 1-2 & 1-4, 2-1 and 2-4 etc.
The 'sine rule' states that (in a right triangle)
a / SinA = b / SinB = c / SinC
So for triangle 1-2-4, although you don't know lengths 1-4 and 2-4, you do know the ratio 1-4 : 2-4. Similarly you know the ratios 2-4 : 3-4 and 1-4 : 3-4 in the other triangles.
I'll leave you to solve point 4. Once you have this point, you can easily solve the z component of 4 using pythagoras' - you'll have the sides 1-4, 1-5 and the length 4-5 will be the z component.
I'll initially assume you know the distances between all pairs of points.
As you say, you can choose one point (A) as the origin, orient a second point (B) along the x-axis, and place a third point (C) along the xy-plane. You can solve for the coordinates of C as follows:
given: distances ab, ac, bc
assume
A = (0,0)
B = (ab,0)
C = (x,y) <- solve for x and y, where:
ac^2 = (A-C)^2 = (0-x)^2 + (0-y)^2 = x^2 + y^2
bc^2 = (B-C)^2 = (ab-x)^2 + (0-y)^2 = ab^2 - 2*ab*x + x^2 + y^2
-> bc^2 - ac^2 = ab^2 - 2*ab*x
-> x = (ab^2 + ac^2 - bc^2)/2*ab
-> y = +/- sqrt(ac^2 - x^2)
For this to work accurately, you will want to avoid cases where the points {A,B,C} are in a straight line, or close to it.
Solving for additional points in 3-space is similar -- you can expand the Pythagorean formula for the distance, cancel the quadratic elements, and solve the resulting linear system. However, this does not directly help you with your steps 2 and 3...
Unfortunately, I don't know a well-behaved exact solution for steps 2 and 3, either. Your overall problem will generally be both over-constrained (due to conflicting noisy distances) and under-constrained (due to missing distances).
You could try an iterative solver: start with a random placement of all your points, compare the current distances with the given ones, and use that to adjust your points in such a way as to improve the match. This is an optimization technique, so I would look up books on numerical optimization.
If you know the distance between the nodes (fixed part of system) and the distance to the tag (mobile) you can use trilateration to find the x,y postion.
I have done this using the Nanotron radio modules which have a ranging capability.

General formula to calculate Polyhedron volume

Given a list of vertices (v), and a list of edges connecting the vertices (e), and a list of surfaces that connect the edges (s), how to calculate the volume of the Polyhedron?
Take the polygons and break them into triangles.
Consider the tetrahedron formed by each triangle and an arbitrary point (the origin).
Sum the signed volumes of these tetrahedra.
Notes:
This will only work if you can keep a consistent CW or CCW order to the triangles as viewed from the outside.
The signed volume of the tetrahedron is equal to 1/6 the determinant of the following matrix:
[ x1 x2 x3 x4 ]
[ y1 y2 y3 y4 ]
[ z1 z2 z3 z4 ]
[ 1 1 1 1 ]
where the columns are the homogeneous coordinates of the verticies (x,y,z,1).
It works even if the shape does not enclose the origin by subracting off that volume as well as adding it in, but that depends on having a consistent ordering.
If you can't preserve the order you can still find some way to break it into tetrahedrons and sum 1/6 absolute value of the determinant of each one.
Edit:
I'd like to add that for triangle mesh where one vertex (say V4) of the tetrahedron is (0,0,0) the determinante of the 4x4 matrix can be simplified to the upper left 3x3 (expansion along the 0,0,0,1 column) and that can be simplified to Vol = V1xV2.V3 where "x" is cross product and "." is dot product. So compute that expression for every triangle, sum those volumes and divide by 6.
Similarly with a polygon where we can split it into triangles and sum the areas,
you could split a polyhedron into pyramids and sum their volumes. But I'm not sure how hard is to implement an algorithm for that.
(I believe there is a mathematical way/formula, like using vectors and matrices.
I suggest to post your question also on http://mathoverflow.net)
I have done this before, but the surface mesh I used always had triangular facets. If your mesh has non triangular facets, you can easily break them up into triangular facets first. Then I fed it to TetGen to obtain a tetrahedralization of the interior. Finally, I added up all the volumes of the tetrahedra. TetGen is reasonably easy to use, and is the only library other than CGAL I know of that can handle complicated meshes. CGAL is pretty easy to use if you don't mind installing a gigantic library and use templates like crazy.
First, break every face into triangles by drawing in new edges.
Now look at one triangle, and suppose it's on the "upper" surface (some of these details will turn out to be unimportant later). Look at the volume below the triangle, down to some horizontal plane below the polyhedron. If {h1, h2, h3} are the heights of the three points, and A is the area of the base, then the volume of the solid will be A(h1+h2+h3)/3. Now we have to add up the volumes of these solids for the upper faces, and subtract them for the lower faces to get the volume of the polyhedron.
Play with the algebra and you'll see that the height of the polyhedron above the horizontal plane doesn't matter. The plane can be above the polyhedron, or pass through it, and the result will still be correct.
So what we need is (1) a way to calculate the area of the base, and (2) a way to tell an "upper" face from a "lower" one. The first is easy if you have the Cartesian coordinates of the points, the second is easy if the points are ordered, and you can combine them and kill two birds with one stone. Suppose for each face you have a list of its corners, in counter-clockwise order. Then the projection of those points on the x-y plane will be counterclockwise for an upper face and clockwise for a lower one. If you use this method to calculate the area of the base, it will come up positive for an upper face and negative for a lower one, so you can add them all together and have the answer.
So how do you get the ordered lists of corners? Start with one triangle, pick an ordering, and for each edge the neighbor that shares that edge should list those two points in the opposite order. Move from neighbor to neighbor until you have a list for every triangle. If the volume of the polyhedron comes up negative, just multiply by -1 (it means you chose the wrong ordering for that first triangle, and the polyhedron was inside-out).
EDIT:
I forgot the best part! If you check the algebra for adding up these volumes, you'll see that a lot of terms cancel out, especially when combining triangles back into the original faces. I haven't worked this out in detail, but it looks as if the final result could be a surprisingly simple function.
Here's a potential implementation for that in Python.
Can anyone please check if it's correct?
I believe that I am missing permutations of the points because my second test (cube) gives 0.666 and not 1. Ideas anyone?
Cheers
EL
class Simplex(object):
'''
Simplex
'''
def __init__(self,coordinates):
'''
Constructor
'''
if not len(coordinates) == 4:
raise RuntimeError('You must provide only 4 coordinates!')
self.coordinates = coordinates
def volume(self):
'''
volume: Return volume of simplex. Formula from http://de.wikipedia.org/wiki/Tetraeder
'''
import numpy
vA = numpy.array(self.coordinates[1]) - numpy.array(self.coordinates[0])
vB = numpy.array(self.coordinates[2]) - numpy.array(self.coordinates[0])
vC = numpy.array(self.coordinates[3]) - numpy.array(self.coordinates[0])
return numpy.abs(numpy.dot(numpy.cross(vA,vB),vC)) / 6.0
class Polyeder(object):
def __init__(self,coordinates):
'''
Constructor
'''
if len(coordinates) < 4:
raise RuntimeError('You must provide at least 4 coordinates!')
self.coordinates = coordinates
def volume(self):
pivotCoordinate = self.coordinates[0]
volumeSum = 0
for i in xrange(1,len(self.coordinates)-3):
newCoordinates = [pivotCoordinate]
for j in xrange(i,i+3):
newCoordinates.append(self.coordinates[j])
simplex = Simplex(newCoordinates)
volumeSum += simplex.volume()
return volumeSum
coords = []
coords.append([0,0,0])
coords.append([1,0,0])
coords.append([0,1,0])
coords.append([0,0,1])
s = Simplex(coords)
print s.volume()
coords.append([0,1,1])
coords.append([1,0,1])
coords.append([1,1,0])
coords.append([1,1,1])
p = Polyeder(coords)
print p.volume()

Resources