Reorganize and commit all the modified TACLeBench code and run scripts

1 files changed, 2014 insertions, 0 deletions

diff --git a/all_pairs/source/susan/susan.c b/all_pairs/source/susan/susan.c
new file mode 100644
index 0000000..fb990b1
--- /dev/null
+++ b/all_pairs/source/susan/susan.c
@@ -0,0 +1,2014 @@
+/**********************************************************************\
+
+  SUSAN Version 2l by Stephen Smith
+  Oxford Centre for Functional Magnetic Resonance Imaging of the Brain,
+  Department of Clinical Neurology, Oxford University, Oxford, UK
+  (Previously in Computer Vision and Image Processing Group - now
+  Computer Vision and Electro Optics Group - DERA Chertsey, UK)
+  Email:    steve@fmrib.ox.ac.uk
+  WWW:      http://www.fmrib.ox.ac.uk/~steve
+
+  (C) Crown Copyright (1995-1999), Defence Evaluation and Research Agency,
+  Farnborough, Hampshire, GU14 6TD, UK
+  DERA WWW site:
+  http://www.dera.gov.uk/
+  DERA Computer Vision and Electro Optics Group WWW site:
+  http://www.dera.gov.uk/imageprocessing/dera/group_home.html
+  DERA Computer Vision and Electro Optics Group point of contact:
+  Dr. John Savage, jtsavage@dera.gov.uk, +44 1344 633203
+
+  A UK patent has been granted: "Method for digitally processing
+  images to determine the position of edges and/or corners therein for
+  guidance of unmanned vehicle", UK Patent 2272285. Proprietor:
+  Secretary of State for Defence, UK. 15 January 1997
+
+  This code is issued for research purposes only and remains the
+  property of the UK Secretary of State for Defence. This code must
+  not be passed on without this header information being kept
+  intact. This code must not be sold.
+
+\**********************************************************************/
+
+/**********************************************************************\
+
+  SUSAN Version 2l
+  SUSAN = Smallest Univalue Segment Assimilating Nucleus
+
+  Email:    steve@fmrib.ox.ac.uk
+  WWW:      http://www.fmrib.ox.ac.uk/~steve
+
+  Related paper:
+  @article{Smith97,
+        author = "Smith, S.M. and Brady, J.M.",
+        title = "{SUSAN} - A New Approach to Low Level Image Processing",
+        journal = "Int. Journal of Computer Vision",
+        pages = "45--78",
+        volume = "23",
+        number = "1",
+        month = "May",
+        year = 1997}
+
+  To be registered for automatic (bug) updates of SUSAN, send an email.
+
+  Compile with:
+  gcc -O4 -o susan susan2l.c -lm
+
+  See following section for different machine information. Please
+  report any bugs (and fixes). There are a few optional changes that
+  can be made in the "defines" section which follows shortly.
+
+  Usage: type "susan" to get usage. Only PGM format files can be input
+  and output. Utilities such as the netpbm package and XV can be used
+  to convert to and from other formats. Any size of image can be
+  processed.
+
+  This code is written using an emacs folding mode, making moving
+  around the different sections very easy. This is why there are
+  various marks within comments and why comments are indented.
+
+
+  SUSAN QUICK:
+
+  This version of the SUSAN corner finder does not do all the
+  false-corner suppression and thus is faster and produced some false
+  positives, particularly on strong edges. However, because there are
+  less stages involving thresholds etc., the corners that are
+  correctly reported are usually more stable than those reported with
+  the full algorithm. Thus I recommend at least TRYING this algorithm
+  for applications where stability is important, e.g., tracking.
+
+  THRESHOLDS:
+
+  There are two thresholds which can be set at run-time. These are the
+  brightness threshold (t) and the distance threshold (d).
+
+  SPATIAL CONTROL: d
+
+  In SUSAN smoothing d controls the size of the Gaussian mask; its
+  default is 4.0. Increasing d gives more smoothing. In edge finding,
+  a fixed flat mask is used, either 37 pixels arranged in a "circle"
+  (default), or a 3 by 3 mask which gives finer detail. In corner
+  finding, only the larger 37 pixel mask is used; d is not
+  variable. In smoothing, the flat 3 by 3 mask can be used instead of
+  a larger Gaussian mask; this gives low smoothing and fast operation.
+
+  BRIGHTNESS CONTROL: t
+
+  In all three algorithms, t can be varied (default=20); this is the
+  main threshold to be varied. It determines the maximum difference in
+  greylevels between two pixels which allows them to be considered
+  part of the same "region" in the image. Thus it can be reduced to
+  give more edges or corners, i.e. to be more sensitive, and vice
+  versa. In smoothing, reducing t gives less smoothing, and vice
+  versa. Set t=10 for the test image available from the SUSAN web
+  page.
+
+  ITERATIONS:
+
+  With SUSAN smoothing, more smoothing can also be obtained by
+  iterating the algorithm several times. This has a different effect
+  from varying d or t.
+
+  FIXED MASKS:
+
+  37 pixel mask:    ooo       3 by 3 mask:  ooo
+                   ooooo                    ooo
+                  ooooooo                   ooo
+                  ooooooo
+                  ooooooo
+                   ooooo
+                    ooo
+
+  CORNER ATTRIBUTES dx, dy and I
+  (Only read this if you are interested in the C implementation or in
+  using corner attributes, e.g., for corner matching)
+
+  Corners reported in the corner list have attributes associated with
+  them as well as positions. This is useful, for example, when
+  attempting to match corners from one image to another, as these
+  attributes can often be fairly unchanged between images. The
+  attributes are dx, dy and I. I is the value of image brightness at
+  the position of the corner. In the case of susan_corners_quick, dx
+  and dy are the first order derivatives (differentials) of the image
+  brightness in the x and y directions respectively, at the position
+  of the corner. In the case of normal susan corner finding, dx and dy
+  are scaled versions of the position of the centre of gravity of the
+  USAN with respect to the centre pixel (nucleus).
+
+  BRIGHTNESS FUNCTION LUT IMPLEMENTATION:
+  (Only read this if you are interested in the C implementation)
+
+  The SUSAN brightness function is implemented as a LUT
+  (Look-Up-Table) for speed. The resulting pointer-based code is a
+  little hard to follow, so here is a brief explanation. In
+  setup_brightness_lut() the LUT is setup. This mallocs enough space
+  for *bp and then repositions the pointer to the centre of the
+  malloced space. The SUSAN function e^-(x^6) or e^-(x^2) is
+  calculated and converted to a uchar in the range 0-100, for all
+  possible image brightness differences (including negative
+  ones). Thus bp[23] is the output for a brightness difference of 23
+  greylevels. In the SUSAN algorithms this LUT is used as follows:
+
+  p=in + (i-3)*x_size + j - 1;
+  p points to the first image pixel in the circular mask surrounding
+  point (x,y).
+
+  cp=bp + in[i*x_size+j];
+  cp points to a position in the LUT corresponding to the brightness
+  of the centre pixel (x,y).
+
+  now for every pixel within the mask surrounding (x,y),
+  n+=*(cp-*p++);
+  the brightness difference function is found by moving the cp pointer
+  down by an amount equal to the value of the pixel pointed to by p,
+  thus subtracting the two brightness values and performing the
+  exponential function. This value is added to n, the running USAN
+  area.
+
+  in SUSAN smoothing, the variable height mask is implemented by
+  multiplying the above by the moving mask pointer, reset for each new
+  centre pixel.
+  tmp = *dpt++ * *(cp-brightness);
+
+\**********************************************************************/
+
+/**********************************************************************\
+
+  Success has been reported with the following:
+
+  MACHINE  OS         COMPILER
+
+  Sun      4.1.4      bundled C, gcc
+
+  Next
+
+  SGI      IRIX       SGI cc
+
+  DEC      Unix V3.2+ 
+
+  IBM RISC AIX        gcc
+
+  PC                  Borland 5.0
+
+  PC       Linux      gcc-2.6.3
+
+  PC       Win32      Visual C++ 4.0 (Console Application)