Matlab m-…le for Digitizing Graphs

Harald E. Krogstad, NTNU, Trondheim

The following simple m-…le helps digitizing graphs and drawings by picking a series of points from the graph
using the mouse pointer provided by the Matlab function ginput (see the Matlab documentation). The
program requires that the user speci…es the x-axis and the y-axis by de…ning two di¤erent points on each.
The axes may be chosen at will and need not cross in a particular point, e.g. the origin. The code is also
able to handle logarithmic axes.
There are two di¤erent coordinate systems to consider:

1. The coordinate system given on the graph itself, which is also the system we want to relate our data
points to
2. The coordinate system of the image of the graph (the pixel numbering used by Matlab)

N
The output from the routine is an N 2 array of data points (xi ; yi )i=1 given in the graph coordinate system.
If the graph only exists on paper, it …rst needs to be scanned in a graphical format Matlab understands, e.g.
the bmp, jpg or png formats (See the Matlab documentation for a survey of other available formats).
With the image of the graph available, the …rst step then consists of getting the graph into Matlab, and this
is done by a call to the Matlab function imread, e.g.
A = imread(’Figure.png’,’png’);
The image is then displayed by calling image,
image(A);
At this stage it may be suitable to zoom and scale the image to a size which is easy to digitize accurately.
The digitizing function is then called (for the …rst time without any input argument):
[data,calib] = digitization([calibin])
Here calib and calibin are Matlab structures that contain the reference calibration points on each axis if
it turns out necessary to carry out the digitization in separate sessions.
Note: If the graph already exists in the form of a Matlab (vector graphics) …g-…le, data values may be
recovered directly by identifying the line graphical elements making up the graph (See Matlab documentation
for further information). As an example, for the current graph, the lines are identi…ed by:
lines = get(gca,’children’)
followed by
xdat = get(lines(1),’xdata’);
ydat = get(lines(1),’ydata’);
... and so on for all lines on the graph.

1 The Algorithm
The digitalization algorithm does not require that the axes on the image are completely orthogonal, or exactly
vertical/horizontal. The coordinates xc and yc are obtained from the data point by skew projections, as
illustrated on Figure 1. This assumes that the distortion is uniform over the whole drawing, which should
be reasonable, at least for scanned images. The points (x0 ; x1 ) and (y0 ; y1 ) are user-de…ned points on the
x- and y-axis with given image as well as real coordinates. The points should be selected at some distance
so the axes are accurately de…ned.
By denoting the data point xd , we obtain the following equations in the 4 variables (t; s; ; ):

1
 y-axis
y1

yc
Data point

y0

x0
x-axis
xc
x1

Figure 1: The graph illustrates how the x and y coordinates for the data point, xd are identi…ed, assuming
straight axes and a pair of points with known locations on both.

xc = x0 + t (x1 x0 ) ;
yc = y0 + s (y1 y0 ) ;
yc = xd (x1 x0 ) ; (1)
xc = xd (y1 y0 ) ;

or

x0 + t (x1 x0 ) = xd (y1 y0 ) ;
xd (x1 x0 ) = y0 + s (y1 y0 ) : (2)

The constants t and s are what is needed to obtain the actual coordinates of the data points. E.g., the
x-coordinates,
xc = x0 + t (x1 x0 ) ; (3)
for a linear axis, and
xc = exp [log x0 + t (log x1 log x0 )] (4)
for a logarithmic axis.
The equations may be written as the matrix system

t
M = R; (5)
s

where

M = [(x1 x0 ) (y1 y0 )] ;
R = [(Xd x0 ) (Xd y0 )] : (6)

It is convenient to introduce corresponding row vectors for (t; s; ; ) and solve

T A
M = R: (7)
B S
The m-…le digitization.m implements the algorithm and also contains an interactive section for de…ning
reference points on the axes.

2
function [data,calib] = digitization(calibin)
% FUNCTION FOR DIGITIZING GRAPHS.
% Axes reference information may be predefined in the structure
% calibin, or it may be created, in which case the user is
% inquired about the following input:
%
% Two points on the x-axis and two points on the y-axis have
% to be chosen, preferably at some distance so as to define
% the axis accurately.
%
% Axes do not need to cross at a particular point and both
% linear and logarithmic axes are accepted.
%
% Harald E. Krogstad, NTNU, 2006
%-------------------------------------------------------------------
if nargin == 0
%
% Dialog for defining the x-axis
x0 = input('Enter x coordinate for x0: ');
disp('Point to x0!')
x0p = (ginput(1))';
%
x1 = input('Enter x coordinate for x1: ');
disp('Point to x1!')
x1p = (ginput(1))';
linlogx = input('Linear/logarithmic x-axis? (0=lin.,1=log.):');
%
% Dialog for defining the y-axis
y0 = input('Enter y coordinate for y0: ');
disp('Point to y0!')
y0p = (ginput(1))';
%
y1 = input('Enter y coordinate for y1: ');
disp('Point to y1!')
y1p = (ginput(1))';
linlogy = input('Linear/logarithmic y-axis? (0=lin.,1=log.):');
%
% Collect axis reference structure
calib.x0 = x0; calib.x1 = x1;
calib.y0 = y0; calib.y1 = y1 ;
calib.x0p = x0p; calib.x1p = x1p;
calib.y0p = y0p; calib.y1p = y1p;
calib.linlogx = linlogx ;
calib.linlogy = linlogy ;
else
%
% Unpack existing axis reference structure
calib = calibin;
x0 = calib.x0; x1 = calib.x1;
y0 = calib.y0; y1 = calib.y1;
x0p = calib.x0p; x1p = calib.x1p;
y0p = calib.y0p; y1p = calib.y1p;
linlogx = calib.linlogx;
linlogy = calib.linlogy;
end
%
% Get data-points
disp('Pick data points, - end with Enter: ')
Xp = ginput;
%
% Find picture coordinates
Ndata = size(Xp(:,1));
Mm = [ (x1p - x0p) (y1p - y0p) ]^(-1);
Xpmx0 = Xp' - x0p*ones(1,Ndata);
Xpmy0 = Xp' - y0p*ones(1,Ndata);
T = Xpmx0'*Mm(1,:)' ;
S = Xpmy0'*Mm(2,:)' ;
%
% Transform to graph (actual) coordinates
if linlogx == 1
% Log x-axis
xdata = exp( log(x0) + (log(x1) - log(x0))*T ) ;
else
% Linear x-axis
xdata = x0 + (x1 - x0)*T;
end
%
if linlogy == 1
% Log y-axis
ydata = exp( log(y0) + (log(y1) - log(y0))*S ) ;
else
% Linear y-axis
ydata = y0 + (y1 - y0)*S;
end
data = [xdata ydata];