-
Notifications
You must be signed in to change notification settings - Fork 0
/
vecFldInterpAnisoB.m
137 lines (122 loc) · 6.14 KB
/
vecFldInterpAnisoB.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
function [interField,prevFlowVecList]=vecFldInterpAnisoB(M, tgtPoints, sigmaU, sigmaV, maxIteration, polygon)
% vecFldInterpAnisoB computes the vector field on a user-specified point set based on vectors
% on a given point set. Initial direction of the flow vector at each given
% point is computed using an isotropic filter. Then this initial vector is
% used for anisotropic filtering at this point. This process is iterated
% until either convergence (in 2-norm) is reached or the maximum number
% iteration is reached. This function do not require users to provide
% initial external flow field. However, if this is desired, function
% "vecFldInterpAnisoA" can be used.
%
% Interpolation is based on an anisotropic Gaussain probability density function of the following form:
%
% p(u, v) = 1 / (2 * pi * sigmaU * sigmaV) * exp(-0.5 * (u / sigmaU)^2 -
% 0.5 * (v / sigmaV)^2);
%
% SYNOPSIS interpFiled = vecFldInterpAnisoB(M, tgtPoints, sigmaU, sigmaV, maxIteration, polygon)
%
% INPUT M : known vector field, stored in a (nx4)-matrix of the form [y0 x0 y x]n
% (where (y0,x0) is the base and (y,x) is the tip of
% the vector).
%
% tgtPoints : points at which vector field needs to be calculated.
% It is stored in a (mx2)-matrix of the form [yg xg]m.In particular,
% this point set can be the same as the given point set in M.
%
% sigmaU, sigmaV : Gauss distribution parameters for calculating
% the probability function. Currently, it is
% assumed that they are contant over the entire
% field. sigmaU is defined along the flow
% direction. sigmaV is defined perpendicular to
% the flow direction.
% NOTICE: these two parameters will also be used to determine a search region, based
% essentially on the three-sigma rule.
%
% maxIteration : maximum number of iterations. This number may not be reached as the calculation
% of flow vector often converges faster.
%
% polygon : (optional - pass polygon=[] to disable). The interpolated vector
% can be cropped to remove vectors outside a given region of interest.
% To create the polygon use the functions ROIPOLY or
% GETLINE. These functions return the polygon vertices
% stored in two vectors y and x. Set polygon=[y x] to
% use with vectorFieldInterp.
%
% OUTPUT interField : interpolated vector field for the points given in tgtPoints.
% DEPENDENCES This function calls vecFldInterpAnisoB
%
% REMARKS This implementation is based on the function and
% interface of "vectorFieldInterp" created by Aaron Ponti,
% 11/18/2002.
%
% AUTHOR Ge Yang
% Laboratory for Computational Cell Biology
% The Scripps Research Institute
%
% DATE OF CREATION March 10, 2004
%
% CHANGES
% ver 0.3: The code is reorganized for better
% readability and initial release.
%
%
% ver 0.4: use MEX function to compute distance
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Vector base positions
srcPoints = M(:, 1:2); % These will be the centers of the probability ellipsoids.
V=[M(:, 3) - M(:, 1) M(:, 4) - M(:, 2)];
SCALEFACTOR = 3; % use 3-sigma rule
[srcLength temp0] = size(srcPoints);
[tgtLength temp0] = size(tgtPoints);
r = 9 * (sigmaU^2 + sigmaV^2); % choose a relatively large one for strong smoothing effect
sigma = sigmaU^2 + sigmaV^2; % still need a Gauss kernel, though isotropic in this case.
% Step 1: Generate the initial flow field using an isotropic filter.
% First, call the MEX function to compute distance. This will save
% computation time.
%size(srcPoints)
%size(tgtPoints)
D = createDistanceMatrix(srcPoints, tgtPoints);
%fprintf('Distance matrix computation done.\n');
prevFlowVecList = zeros(tgtLength, 2);
for i = 1 : tgtLength
weight_sum = 0;
vecsum = [0 0];
for j = 1 : srcLength
temp = D(j, i); % the Euclidean distance
if temp <= r % actually comparing distance square
weight = exp(-0.5 * temp / sigma); % will normalize later. So no need to compute leading coefficient
vecsum = vecsum + weight * V(j, :);
weight_sum = weight_sum + weight;
end
end
prevFlowVecList(i, :) = vecsum / weight_sum; % This is our initial vector field
end
clear D;
% Step 2: Call vecFldInterpAnisoA iterative to compute the vector field
error = 10^6; % just give a large number. value not meaningful
threshold = 1e-2;
%fprintf('Initial vector field computation done.\n');
i = 0;
while (1)
newFlowVecList = vecFldInterpAnisoA(M, tgtPoints, prevFlowVecList, sigmaU, sigmaV);
if size(M,1) == 0
newFlowVecList = zeros(size(prevFlowVecList));
end
error = sqrt(sum(newFlowVecList - prevFlowVecList).^2) / tgtLength;
if (error < threshold)
interField = newFlowVecList;
% fprintf('Computation converged in flow field calculation. Now return to the calling point.\r');
return;
else
i = i + 1;
if (i >= maxIteration)
% fprintf('WARNING: Maximum number reached in iterative calculation of vector field. Now return to the calling point. \n');
interField = newFlowVecList;
return;
end
end
%fprintf('Flow field calculation: iteration # %d\n', i);
prevFlowVecList = newFlowVecList;
end