FASTCPDA - Fast corner detection.

Description
Algorithm
Syntax
Input
Property [propertyname propertyvalues]
Outputs
References
Acknowledgement
See also
Function implementation
Subfunction

Description

Fast corner detector based on the chord-to-point distance accumulation technique of [AL08,ALFR09], itself derived from the curvature-based approach introduced in [HY04,HY08].

Algorithm

The original algorithm by [HY04,HY08] works by the following steps:

extract the edge contours from the edge-map, fill the gaps in the contours.
compute curvature at a low scale for each contour to retain all true corners.
all of the curvature local maxima are considered as corner candidates, then rounded corners and false corners due to boundary noise and details were eliminated.
end points of line mode curve were added as corner, if they are not close to the above detected corners.

Improvements due to CPDA (Chord-to-Point Distance Accumulation) rely on the observation that setting different sigma values based on the curve length is rather impractical, since a different set of edges may be extracted from the test images and the extracted edges often vary in lengths. Therefore, sigma is set to sigma=3 for all curves for the CPDA detector. This is different from [AG08] where sigma was set one of three values 1, 2 and 3 based on the curve-length.

Syntax

   [ptcorner,cornermap] = FASTCPDA(I);
   [ptcorner,cornermap] = FASTCPDA(I, 'Property', propertyvalue, ...);
   [ptcorner,cornermap] = FASTCPDA(emap);
   [ptcorner,cornermap] = FASTCPDA(emap, 'Property', propertyvalue, ...);

Input

V : input image (see also variables 'der' and 'sig' below) to analyze or (binary/logical) map of already detected edges.

Property [propertyname propertyvalues]

'gap' : maximum gap between the successive edge-points, used to fill the gaps in the contours; if the gap between the sucessive points is more than Gap_size, then they would be in two different edges; default: gap=1 pixel.

'end' : flag to indicate whether the end-points of a curve are detected as corners (true) or not; default: end=false.

'thang' : threshold on angle orientation; it denotes the maximum obtuse angle that a corner can have when it is detected as a true corner; default: thang=157;

'der' : in the case, the input image is to be processed, this optional string stores the method used for calculating the edge map: 'matlab', 'vista' or 'kovesi'; default: der='vista'.

'sig' : similarly, this (optional) variable gives standard deviation of the Gaussian filter used for smoothing the image and computing the curvature; default: sig=1.

'reduce' : logical value or string defining the way the different channels of a multispectral image are combined into the output edge map:

'igray' : the input RGB image is converted to a gray image using the function RGB2GRAY,
'imax', 'isum' : the input image (any dimension) is converted to a gray image by taking the sum and the max over the different channels resp.,
'gmax' : gradients are computed for the different channels and their local pixelwise max is given as a single input to the edge detector;
'eor' : calculations are made like for a multispectral image (as if reduce=false), but the final edge map is taken as the logical OR of the output edge maps of the different channels;

in true case, it is set to 'sum'; default: reduce=false, ie. no 'combination' is used, a multichannel map is output; this parameter is naturally ignored when I is a scalar image.

Outputs

cout : (n,2) matrix containing the coordinates of the detected corners, where n is the number of detected corners.

cd : cpda curvature values of the detected corners.

References

[MS98] F. Mokhtarian and R. Suomela: "Robust image corner detection through curvature scale space", IEEE Trans. on Pattern Analysis and Machine Intelligence, 20(12):1376-1381, 1998 http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=735812

[HY04] X.C. He and N.H.C. Yung: "Curvature scale space corner detector with adaptive threshold and dynamic region of support", Proc. ICPR, vol. 2, pp. 791-794, 2004. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1334377

[HY08] X.C. He and N.H.C. Yung: "Corner detector based on global and local curvature properties", Optical Engineering, 47(5):057008, 2008. http://spiedigitallibrary.org/oe/resource/1/opegar/v47/i5/p057008_s1

[AL08] M. Awrangjeb and G. Lu: "Robust image corner detection based on the chord-to-point distance accumulation technique", IEEE Trans. on Multimedia, 10(6):1059-1072, 2008. http://www.gscit.monash.edu.au/gscitweb/loid.php?loid=905299&mimetype=application/pdf

[ALFR09] M. Awrangjeb, G. Lu, C.S. Fraser and M. Ravanbakhsh: "“A Fast Corner Detector Based on the Chord-to-Point Distance Accumulation Technique", Proc. DICTA, pp. 519-525, 2009. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5384897

Acknowledgement

This is an adaptation of the original code of [AL08,ALFR09] available at: http://www.mathworks.com/matlabcentral/fileexchange/28207-a-fast-corner-detector-based-on-the-chord-to-point-distance-accumulation-technique

Part of this code was already from the source code of [HY04,HY08] available at: http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=7652&objectType=file

Function implementation

function [ptcorner,cornermap] = fastcpda(V, varargin)

parsing and checking parameters

error(nargchk(1, 21, nargin, 'struct'));
error(nargoutchk(1, 2, nargout, 'struct'));

% mandatory parameter
if ~(isnumeric(V) || islogical(V))
    error('fastcpda:inputerror','numeric or logical matrix required in input');
end

p = createParser('FASTCPDA');   % create an instance of the inputParser class.
% optional parameters
p.addParamValue('thang',157, @(x)x>=0 && x<=180);
p.addParamValue('gap',1, @(x)isscalar(x) && x>=1);
p.addParamValue('end',false,@(x)islogical(x));
% used in case an edge map has to be produced first
p.addParamValue('sig', 1, @(x)isscalar(x) && x>=0.05);
p.addParamValue('der', 'vista', @(x)ischar(x) && ...
    any(strcmpi(x,{'matlab','vista','kovesi'})));
p.addParamValue('reduce', false, @(x)islogical(x) || ...
    (ischar(x) && any(strcmpi(x,{'igray','imax','isum','gmax','eor'}))));

% parse and validate all input arguments
p.parse(varargin{:});
p = getvarParser(p);

calculation

if ~islogical(V)
    V = cannyedge_base(V, p.sig, p.der, [], p.reduce);
end

[ptcorner, cornermap] = fastcpda_base(V, p.gap, p.thang, p.end);

display

if p.disp
    tmp = V; % to show corners on the input image
    for c=1:size(ptcorner,1)
        for i=1:size(ptcorner{c},1)
            tmp(:,:,c) = mark(tmp(:,:,c),ptcorner{c}(i,1),ptcorner{c}(i,2),3);
        end
    end
    figure, imagesc(tmp), axis image off, title('fast CPDA corner map');
    if size(tmp,3) == 1,   colormap gray;  end
end

end % end of fastcpda

Subfunction

MARK - Show corners into the output images or into the edge-image.

%--------------------------------------------------------------------------
function I1 = mark(I, x, y, w)
[M,N,~] = size(I);

I1 = I;

if isa(I,'logical')
    I1(max(1,x-floor(w/2)):min(M,x+floor(w/2)),max(1,y-floor(w/2)):min(N,y+floor(w/2)),:) = ...
        (I1(max(1,x-floor(w/2)):min(M,x+floor(w/2)),max(1,y-floor(w/2)):min(N,y+floor(w/2)),:)<1);
    I1(x-floor(w/2)+1:x+floor(w/2)-1,y-floor(w/2)+1:y+floor(w/2)-1,:) = ...
        I(x-floor(w/2)+1:x+floor(w/2)-1,y-floor(w/2)+1:y+floor(w/2)-1,:);

else
    I1(max(1,x-floor(w/2)):min(M,x+floor(w/2)),max(1,y-floor(w/2)):min(N,y+floor(w/2)),:) = ...
        (I1(max(1,x-floor(w/2)):min(M,x+floor(w/2)),max(1,y-floor(w/2)):min(N,y+floor(w/2)),:)<128)*255;
    I1(x-floor(w/2)+1:x+floor(w/2)-1,y-floor(w/2)+1:y+floor(w/2)-1,:) = ...
        I(x-floor(w/2)+1:x+floor(w/2)-1,y-floor(w/2)+1:y+floor(w/2)-1,:);
end

end % end of mark