First keep in mind that this is not a Matlab tutorial.

This is just a list of

tricks I have found useful while writing my toolboxes available on the Matlab
Central repository
http://www.mathworks.com/matlabcentral/
You can e-mail me if you have corrections about these pieces of code, or if you
would like to add your own tips to those described in this document.
Contents
1 General Programming Tips 1
2 General Mathematical Tips 4
3 Advanced Mathematical Tips 4
4 Signal and Image Processing Tips 6
5 Graph Theory Tips 10
1 General Programming Tips
• Suppress entries in a vector.
x( 3:5 ) = [];
• Reverse a vector.
x = x(end:­1:1);
• Compute the running time of a function call.
1
tic; fft(rand(500)); disp( [’it takes ’ num2str(toc) ’s.’] );
• Make a array full of a
% guess which one is the fastest ?
tic; NaN*ones(2000,2000); toc;
tic; repmat(NaN,2000,2000); toc;
• Turn an nD array into a vector.
x = x(:);
• Compute the maximum value of an nD array.
m = max(x(:));
• Access a matrix from a list of entries. Here, we have I = [I1; I2] and
y(i) = M( I1(i), I2(i) )
J = sub2ind(size(M), I(1,:),I(2,:) );
y = M(J);
• Create a function that take optional arguments in a struct.
function y = f(x,options)
% parse the struct
if nargin<2
options.null = 0; % force creation of options
end
if isfield(options, ’a’)
options.a = 1; % default value
end
a = options.a;
if isfield(options, ’b’)
options.b = 1; % default value
end
b = options.b;
% Here the body of the function ...
• Create a graphical waitbar.
n = 100;
h = waitbar(0,’Waiting ...’);
for i=1:n
waitbar(i/n);
% here perform some stuff
end
close(h);
• How to duplicate a character n times.
2
str = char( zeros(n,1)+’*’ );
• Output a string without carriage return.
fprintf(’Some Text’);
• Assign value v in a nD array at a position ind (lenth-n vector).
ind = num2cell(ind);
x( ind{:} ) = v;
• Save the current figure as an image in e.g. EPS file format.
saveas(gcf, str, ’png’);
• Remove the ticks from a drawing.
set(gca, ’XTick’, []);
set(gca, ’YTick’, []);
• Saving and loading an image.
saveas(gcf, ’my image’, ’png’); % save
M = double( imread( ’my image.png’ ) ); % load
• Saving and loading a matrix M in a binary file.
[n,p] = size(M); % saving
str = ’my file’; % name of the file
fid = fopen(str,’wb’);
if fid<0
error([’error writing to file ’, str]);
end
fwrite(fid,M,’double’);
fclose(fid);
% loading
fid = fopen(str,’rb’);
if fid<0
error([’error reading file ’,str]);
end
[M, cnt] = fread(fid,[n,p],’double’);
fclose(fid);
if cnt =n*p
error([’Error reading file ’, str]);
end
• Find the angle that makes a 2D vector x with the vector [1,0]
% just the angle
theta = atan2(x(2),x(1));
% if you want to compute the full polar decomposition
[theta,r] = cart2pol(x);
3
2 General Mathematical Tips
• Rescale the entries of a vector x so that it spans [0, 1]
m = min(x(:)); M = max(x(:));
x = (b­a) * (x­m)/(M­m) + a;
• Generate n points evenly sampled.
x = 0:1/(n­1):1; % faster than linspace
• Compute the L2 squared norm of a vector or matrix x.
m = sum(x(:).ˆ2);
• Subsample a vector x or an image M by a factor 2.
x = x(1:2:end); % useful for wavelet transform
M = M(1:2:end,1:2:end);
• Compute centered finite differences.
D1 = [x(2:end),x(end)];
D2 = [x(1),x(1:end­1)];
y = (D1­D2)/2;
• Compute the prime number just before n
n = 150;
P = primes(n); n = P(end);
• Compute J, the reverse of a permutation I, i.e. an array which contains the
number 1:n in arbitrary order.
J(I) = 1:length(I);
• Shuffle an array x.
y = x( randperm(length(x)) );
3 Advanced Mathematical Tips
• Generate n points x sampled uniformly at random on a sphere.
4
% tensor product gaussian is isotropic
x = randn(3,n);
d = sqrt( x(1,:).ˆ2+x(2,:).ˆ2+x(2,:).ˆ2 );
x(1,:) = x(1,:)./d;
x(2,:) = x(2,:)./d;
x(3,:) = x(3,:)./d;
• Construct a polygon x whose ith sidelength is s(i). Here x(i) is the complex
affix of the ith vertex.
theta = [0;cumsum(s)];
theta = theta/theta(end);
theta = theta(1:(end­1));
x = exp(2i*pi*theta);
L = abs(x(1)­x(2));
x = x*s(1)/L; % rescale the result
• Compute y, the inverse of an integer x modulo a prime p.
% use Bezout thm
[u,y,d] = gcd(x,p);
y = mod(y,p);
• Compute the curvilinear abscise s of a curve c. Here, c(:,i) is the ith point
of the curve.
D = c(:,2:end)­c(:,1:(end­1));
s = zeros(size(c,2),1);
s(2:end) = sqrt( D(1,:).ˆ2 + D(2,:).ˆ2 );
s = cumsum(s);
• Compute the 3D rotation matrix M around an axis v
% taken from the OpenGL red book
v = v/norm(v,’fro’);
S = [0 ­v(3) v(2); v(3) 0 ­v(1); ­v(2) v(1) 0];
M = v*transp(v) + cos(alpha)*(eye(3) ­ v*transp(v)) + sin(alpha)*S;
• Compute a VanderMonde matrix M i.e. M(i,j)=x(i)ˆj for j=0:d.
n = length(x); % first method
[J,I] = meshgrid(0:d,1:n);
A = x(I).ˆJ;
% second method, less elegant but faster
A = ones(n);
for j = 2:n
A(:,j) = x.*A(:,j­1);
end
• Threshold (i.e. set to 0) the entries below T.
5
% first solution
x = (abs(x)>=T) .* x;
% second one : nearly 2 times slower
I = find(abs(x)<T); x(I) = 0;
• Keep only the n biggest coefficients of a signal x (set the others to 0).
[,I] = sort(abs(x(:))); x( I(1:end­n) ) = 0;
• Draw a 3D sphere.
p = 20; % precision
t = 0:1/(p­1):1;
[th,ph] = meshgrid( t*pi,t*2*pi );
x = cos(th);
y = sin(th).*cos(ph);
z = sin(th).*sin(ph);
surf(x,y,z, z.*0);
% some pretty rendering options
shading interp; lighting gouraud;
camlight infinite; axis square; axis off;
• Project 3D points on a 2D plane (best fit plane). P(:,k) is the kth point.
for i=1:3 % substract mean
P(i,:) = P(i,:) ­ mean(P(i,:));
end
C = P*P’; % covariance matrix
% project on the two most important eigenvectors
[V,D] = eigs(C);
Q = V(:,1:2)’*P;
4 Signal and Image Processing Tips
• Compute circular convolution of x and y.
% use the Fourier convolution thm
z = real( ifft( fft(x).*fft(y) ) );
• Display the result of an FFT with the 0 frequency in the middle.
x = peaks(256);
imagesc( real( fftshift( fft2(x) ) ) );
• Resize an image M (new size is (p1,q1)).
[p,q] = size(M); % the original image
[X,Y] = meshgrid( (0:p­1)/(p­1), (0:q­1)/(q­1) );
% new sampling location
[XI,YI] = meshgrid( (0:p1­1)/(p1­1) , (0:q1­1)/(q1­1) );
M1 = interp2( X,Y, M, XI,YI ,’cubic’); % the new image
6
• Build a 1D gaussian filter of variance s.
x = ­1/2:1/(n­1):1/2;
f = exp( ­(x.ˆ2)/(2*sˆ2) );
f = f / sum(sum(f));
• Build a 2D gaussian filter of variance s.
x = ­1/2:1/(n­1):1/2;
[Y,X] = meshgrid(x,x);
f = exp( ­(X.ˆ2+Y.ˆ2)/(2*sˆ2) );
f = f / sum(f(:));
• Perform a 1D convolution of signal f and filter h with symmetric boundary
conditions. The center of the filter is 0 for odd length filter, and 1/2 otherwise
n = length(x); p = length(h);
if mod(p,2)==1
d1 = (p­1)/2; d2 = (p­1)/2;
else
d1 = p/2­1; d2 = p/2;
end
xx = [ x(d1:­1:1); x; x(end:­1:end­d2+1) ];
y = conv(xx,h);
y = y( (2*d1+1):(2*d1+n) );
• Same but for 2D signals
n = length(x); p = length(h);
if mod(p,2)==1
d1 = (p­1)/2; d2 = (p­1)/2;
else
d1 = p/2­1; d2 = p/2;
end
xx = [ x(d1:­1:1,:); x; x(end:­1:end­d2+1,:) ];
xx = [ xx(:,d1:­1:1), xx, xx(:,end:­1:end­d2+1) ];
y = conv2(xx,h);
y = y( (2*d1+1):(2*d1+n), (2*d1+1):(2*d1+n) );
• Extract all 0th level curves from an image M an put these curves into a cell
array c list.
c = contourc(M,[0,0]);
k = 0; p = 1;
while p < size(c, 2) % parse the result
lc = c(2,p); % length of the curve
cc = c(:,(p+1):(p+lc));
p = p+lc+1;
k = k+1;
c list{k} = cc;
end
• Quick computation of the integral y of an image M along a 2D curve c (the
7
curve is assumed in [0, 1]2)
cs = c*(n­1) + 1; % scale to [1,n]
I = round(cs);
J = sub2ind(size(M), I(1,:),I(2,:) );
y = sum(M(J));
• Draw the image of a disk and a square.
n = 100; x = ­1:2/(n­1):1;
[Y,X] = meshgrid(x,x);
c = [0,0]; r = 0.4; % center and radius of the disk
D = (X­c(1)).ˆ2 + (Y­c(2)).ˆ2 < rˆ2;
imagesc(D); % a disk
C = max(abs(X­c(1)),abs(Y­c(2)))<r;
imagesc(C); % a square
• Draw a 2D function whose value z is known only at scattered 2D points
(x,y).
n = 400;
x = rand(n,1); y = rand(n,1);
% this is an example of surface
z = cos(pi*x) .* cos(pi*y);
tri = delaunay(x,y); % build a Delaunay triangulation
trisurf(tri,x,y,z);
• Generate a signal whose regularity is Cα (Sobolev).
alpha = 2; n = 100;
y = randn(n,1); % gaussian noise
fy = fft(y);
fy = fftshift(fy);
% filter the noise with |omega|ˆ­alpha
h = (­n/2+1):(n/2);
h = (abs(h)+1).ˆ(­alpha­0.5);
fy = fy.*h’;
fy = fftshift(fy);
y = real( ifft(fy) );
y = (y­min(y))/(max(y)­min(y));
• Generate a signal whose regularity is nearly Cα−1/2.
alpha = 3; n = 300;
x = rand(n,1); % uniform noise
for i=1:alpha % integrate the noise alpha times
x = cumsum(x ­ mean(x));
end
• Compute the PSNR between to signals x and y.
d = mean( mean( (x­y).ˆ2 ) );
m = max( max(x(:)),max(y(:)) );
PSNR = 10*log10( m/d );
8
• Evaluate a cubic spline at value t (can be a vector).
x = abs(t) ;
I12 = (x>1)&(x<=2); I01 = (x<=1);
y = I01.*( 2/3­x.ˆ2.*(1­x/2) ) + I12.*( 1/6*(2­x).ˆ3 );
• Perform spectral interpolation of a signal x (aka Fourier zero-padding). The
original size is n and the final size is p
n = length(x); n0 = (n­1)/2;
f = fft(x); % forward transform
f = p/n*[f(1:n0+1); zeros(p­n,1); f(n0+2:n)];
x = real( ifft(f) ); % backward transform
• Compute the approximation error err= ||f−fM||/||f|| obtained when keeping
the M best coefficients in an orthogonal basis.
% as an example we take the decomposition in the cosine basis
M = 500;
x = peaks(128); y = dct(x); % a sample function
[tmp,I] = sort(abs(y(:)));
y(I(1:end­M)) = 0;
err = norm(y,’fro’)/norm(x,’fro’); % the relative error
xx = idct(y); imagesc(xx); % the reconstructed function
• Perform a JPEG-like transform of an image x (replace dct by idct to compute
the inverse transform).
bs = 8; % size of the blocks
n = size(x,1); y = zeros(n,n);
nb = n/bs; % n must be a multiple of bs
for i=1:nb
for j=1:nb
xsel = ((i­1)*bs+1):(i*bs);
ysel = ((j­1)*bs+1):(j*bs);
y(xsel,ysel) = dct(x(xsel,ysel));
end
end
• Extract interactively a part MM of an image M.
[n,p] = size(M);
imagesc(M);
axis image; axis off;
sp = getrect;
sp(1) = max(floor(sp(1)),1); % xmin
sp(2) = max(floor(sp(2)),1); % ymin
sp(3) = min(ceil(sp(1)+sp(3)),p); % xmax
sp(4) = min(ceil(sp(2)+sp(4)),n); % ymax
MM = M(sp(2):sp(4), sp(1):sp(3));
9
5 Graph Theory Tips
• Compute the shortest distance between all pair of nodes (D is the weighted
adjacency matrix).
% non connected vectices must have Inf value
N = length(D);
for k=1:N
D = min(D,repmat(D(:,k),[1 N])+repmat(D(k,:),[N 1]));
end
D1 = D;
• Turn a triangulation into an adjacency matrix.
nvert = max(max(face));
nface = length(face);
A = zeros(nvert);
for i=1:nface
A(face(i,1),face(i,2)) = 1; A(face(i,2),face(i,3)) = 1;
A(face(i,3),face(i,1)) = 1;
% make sure that all edges are symmetric
A(face(i,2),face(i,1)) = 1; A(face(i,3),face(i,2)) = 1;
A(face(i,1),face(i,3)) = 1;
end
10