LO-RANSAC mex binaries, compiled for 32 and 64 bit Matlab.

Usage:
  [H, INL] = loransacH(TC_PAIRS, THRESHOLD, LO[, CONF, INL_LIMIT, MAX_SAMPLES])
  [H, INL] = loransacHiter(TC_PAIRS, THRESHOLD, LO[, CONF, INL_LIMIT, MAX_SAMPLES])
or
  [F, INL] = loransacF(TC_PAIRS, THRESHOLD, LO[, CONF, INL_LIMIT, MAX_SAMPLES])
  [F, INL] = loransacFiter(TC_PAIRS, THRESHOLD, LO[, CONF, INL_LIMIT, MAX_SAMPLES])

Robustly estimates homography/fundamental matrix from a set of
correspondences. The quadratic cost function (MSAC's) is used.

Input:

TC_PAIRS 6xN matrix, each column represents homogenous
coordinates of two corresponding points in two images

THRESHOLD is a squared value of Sampson's error for inliers

LO (int32) triggers Local Optimization step (LO > 0)

CONF (default = 0.95) defines the confidence in the solution

MAX_SAMPLES (default = 100000) defines the maximal number of
samples

INL_LIMIT limited number of inliers for the computation of
least squares (set to 0 to turn this limit off, default =
0/OFF).

Output:

H/F 3x3 matrix representation of the geometry constraints

INL 1xN logical array indicating inliers (1) and outliers (0)

model of the epipolar geometry F:
  rank(F)=2 and
  TC_PAIRS(1:3, i)' * F * TC_PAIRS(4:6, i) = 0, for INL(i) == 1
model of the homography H:
  cross (TC_PAIRS(1:3, i), H * TC_PAIRS(4:6, i)) = 0,
  for INL(i) == 1

Procedures (in the paper):
 MSAC:			loransacH/loransacF with LO == 0
 MSAC.LO:		loransacH/loransacF with LO > 0
 MSAC.LOiter:	loransacHiter/loransacFiter with LO > 0
