1.3	A corrupt transform file can no longer drive an absurd allocation.
	API change: vs_vector_set() and vs_vector_set_dup() return
	VS_OK/VS_ERROR and report the replaced element through a
	`void **old' out parameter.
	Relicensed the library (src/) from GPL-2.0-or-later to
	LGPL-2.1-or-later, with the agreement of its copyright holders;
	see RELICENSE.md.
	Removed the transcode plugins and the vendored sse2neon.h shim.
	Transcode is no longer maintained, and ARM has had native NEON
	kernels since the SIMD dispatch rework, so nothing included the
	shim any more.
	L1-optimal camera path (Grundmann et al., CVPR 2011) implemented and
	enabled: camPathAlgo=VSOptimalL1 now really runs the optimization
	instead of falling back to the gaussian filter.  No new config
	fields: the optimization takes its zoom budget from zoom/optZoom
	and its horizon from smoothing.
	Built-in interior point LP solver, no external dependency; GLPK
	selectable with cmake -DVIDSTAB_LPSOLVER=glpk.
1.1     use openMP to do parallel execution
1.0 = 0.98 (just because of API changes the version number was bumped
0.98...
	small measurement fields added that only scan around mean
	diagnostics in show=1 and show=2 use lines now.
	fix zoom detection/treatment
	zoomspeed param for optzoom
	gaussian filter for cam path optimization
0.97
	optzoom = 2: adaptive zooming added
	border bluring for interpolation>=bilinear
	spurious black pixels bug fixed
0.96
	libavfilter plugins work
	flexible support of pixel formats, ala ffmpeg
0.95
	virtual tripod mode

0.94
	local motions are now stored in the file (file format changed,
	 old one still supported)
	BUGFIX for zero-transforms

0.93
	single pass version filter_deshake
	made stabilize part print the transforms to the file right away

0.92
	boxblur added and unsharp filter removed for motiondetection

0.91
	spiral search for matches in compareSubImg

0.90
	complete restructuring into own library
	Fixed point arithmetic used for transform code
	motiondetection optimized with ORC: speedup by factor 4
	ORC 0.4.14 is now required
	added tests for speed and functional tests
	removed biquadratic interpolation routine

0.80
	keep borders with stabilized last frame (as it should be).
	Pointed out by Guido Torelli.

0.79
	speed optimizations using SSE by Alexey Osipov
	search tree cut, spiral search and sse:
	 together speedup factor ~8 of stabilize run
	stepsize is increased stepwise
	keep border at transform plugin improved


0.77
	interpolation routines improved a lot, thanks to hints on ffmpeg list
	added bicubic interpolation that uses 4x4 pixel (useful for large zoom values)
	bilinear interpolation is now the default
	transform plugin uses last transform for the remaining frames
	 -> this enables to use the transform plugin for constant transformations

0.76
	BUGFIX in calcFieldTransYUV caused SEGFAULT

0.75
	two meta parameter introduces: shakiness and accuracy and
	removal of fieldnum, fielsize, maxshift, maxfieldnum
	field placement changed: Now they fill the frame perfectly
	added bluring such that stepsize can be much larger: much faster now!
        linear interpolation is now the default
	uncertain angles are set to 0 (more robustness)

0.70    selects a maximal number of fields according to their contrast
	nicely distributed over the frame
	Todo: remove fieldnum, allowmax (now false), and compressed parameter
	Todo: add accuracy parameter instead of maxfields
        Bugfix of contrast routine (error in the calculation of maxi)

0.62
	different interpolation functions. No interpolation is default and
	 seems fine in most applications.
	field placement changed. Now they fill the frame better.
	"compress" option moves them more to the center (vertically)
	fields and their areas and transforms can be drawn on the frame.

0.6
	new field placement allows for any number of fields
	 and respects aspect ratio
	fields with low contrast are ignored
	non-symmetric field setups are possible

0.5
	global and per-frame zooming supported
	optimal zoom value is calculated to avoid moving border	effect
	add filter unsharp to sharpen the frames
	write parameters to transforms file

0.44
	bugfixing
	finished RGB #support
	transforms file format supports comments

0.42
	input filename and tc_list stuff used
0.41
	code style from transcode used
0.4
	moved to CVS version we use a cleaned median to select the right
	transform from the fields
0.3
	Support RGB (untestet)
	maxshift default 50
	allowmax added
	scan only every second shift and then make a second small
          scan around the best match
	removed black rectangle around transformed frame
