For-loop vectorization

Hello people, I'm sort of a novice matlab user at best. Then, How to effectively vectorize the loop in this code?

se = strel('diamond',2); 
row = size(P, 1);  % where P is a RGB image 
col = size(P, 2);
crist = zeros(row,col, 1);
num = sum(sum(S)); % where S is a binary image (P and S have the same size)
[Y, X]=find(S, num); 
for i = 1:num
    x1 = X(i); 
    y1 = Y(i); 
    d1 = tdist3D(P, x1, y1);
    t1 = d1 < dist; % where dist is a threshold number defined by the user
    crist = or(crist, bwselect(bwmorph(imerode(t1, se), 'close', 2), x1, y1, 4));
end
.....

where, tdist3D is the fallowing function:

function J=tdist3D(A,x,y)
% Distance transformation between the RGB image pixels and
% the (x,y) pixel of the binary image.
% 
% Syntax:
%     J = tdist3D(A,x,y)
%
% Input:
%    A:   image RGB
%    x,y: pixel of the binary image
%
% Output:
%    J:   image distance

temp = ones(size(A));
Ref = cat(3, temp(:, :, 1).*A(y, x, 1), temp(:, :, 1).*A(y, x, 2), temp(:, :, 1).*A(y, x, 3));

D = A - Ref;
D2 = D.^2;
SD2 = sum(D2, 3);
J = SD2.^0.5;

Thank in advance for any help.

JulioC

Answer 1

[Jan]

• sqrt() is much faster than ^0.5.
• When you do not need the 3rd component of temp, do not create it:

temp = ones(size(A, 1), size(A, 2));

• What is the reason to multiply by ONES?! A multiplication with a scalar 1 would be much faster. But I guess you want a converion to a double? Then use double(A).
• The bottlenecks of your program are bwselect, bwmorph and imerode. Therefore a vectorization will not help noticably.