sift.1

linux操作系统下的sift算法

.Dd August 22, 2006
.Os
.Dt SIFT 1
.Sh NAME
.Nm sift
.Nd Scale Invariant Feature Transform
.Sh SYNOPSIS
.Nm
.Op Fl vhb 
.Op Fl o Ar filename
.Op Fl k Ar filename
.Op Fl p Ar prefix
.Op Fl S Ar number 
.Op Fl O Ar number
.Op Fl t Ar number
.Op Fl e Ar number
.Op Fl \-save\-gss
.Op Fl \-no\-descriptors
.Op Fl \-no\-orientations
.Op Fl \-stable\-order
file.pgm ...
.Sh OPTIONS
.Bl -tag
.It Fl \-verbose , Fl v
Be verbose.
.It Fl \-help , Fl h
Print a summary of options.
.It Fl \-output Ar outfile , Fl o Ar outfile
Wirte the keypoints to 
.Ar outfile.
.It Fl \-prefix Ar prefix , Fl p Ar prefix
Derive the output filename prefixing 
.Ar prefix
to the input filename.
.It Fl \-binary , Fl b
Write the descriptors to a binary file. The name of this file
is derived from the output filename by removing the `.key' suffix
and adding the `.desc' suffix.
.It Fl \-keypoints Ar filename, Fl k Ar filename
Instead of running the SIFT detector, read the keypoints from
.Ar filename
(but still compute the descriptors).
.It Fl \-save\-gss
Save the layers of the Gaussian scale space. This produces a
\.pgm file for each level of the pyramid. The name of these file
is derived from the output filename by removing the `.pgm' suffix,
and adding the index of the level and the `.pgm' suffix.
.It Fl \-octaves Ar number ,  Fl O Ar number
Set the number of octave to 
.Ar number.
If not specified, this value is automatically
computed to span all possible octaves.
.It Fl \-levels Ar number ,  Fl S Ar number
Set the number of levels per octave to 
.Ar number.
The default value is 3.
.It Fl \-first\-octave Ar number
Set the index of the first octave of the pyramid to
.Ar number.
The default value is -1.
.It Fl \-threshold Ar number, Fl t Ar number 
Set the SIFT detector threshold to
.Ar number.
.It Fl \-edge\-threshold Ar number , Fl e Ar number
Set the SIFT detector edge rejection threshold to
.Ar number.
.It Fl \-no\-descriptors
Do not compute nor wirte to the output file the descriptors.
.It Fl \-no\-orientations
Set implicitly the orientation of all keypoints to zero.
.It Fl \-stable\-order
Preserve the order of keypoints read from a file.
.El
.Sh OVERVIEW
.Nm
computes D. Lowe's Scale Invariant Feature Transfroms (SIFT). The
program can be used to process either a single image or several images
in a batch.  It is possible to customize the most important parameters
of the algorithm and to generate descriptors for keypoints computed
externally.
.Pp
In the most simple form
.Nm
takes an image in PGM format and computes its SIFT keypoints and the
relative descriptors, producing a
.Ql .key
text file. This file has one line per keypoint, with the x and y
coordiantes (pixels), the scale (pixels), the orientation (radians)
and 128 numbers (in the range 0-255) representing the descriptor. This
file is almost equivalent to the output of D. Lowe's original
implementation, except that x and y are swapped and the orientation is
negated.
.Pp
The SIFT algorithm computes a Gaussian pyrmaid of the input image. To
change the number of octaves, the index of the first octave and the
number of levels per octave of the pyramid, use the options
.Fl \-octaves ,
.Fl \-first-octave
and
.Fl \-levels
respectively. Note that setting
.Fl \-first-octave
to -1, -2, ... will cause the base of the pyramid to be two,
three, ... times larger than the input image.
.Pp
In order to select reliable keypoints, SIFT rejects extrema of the
Difference of Gaussian scale space which are smaller than a threshold
specified by
.Fl \-threshold .
It also rejects keypoints that have an on-edge score above another
threshold specified by
.Fl \-edge-threshold.
.Pp
By default, the name of the ouptut file is obtained by removing the
.Ql .pgm
suffix (if any) from the name of the input and then appending the
.Ql .key
suffix. The option
.Fl \-output Ar name
changes this behaviro and uses the name
.Ar name
instead. 
.Fl \-output
works only when processing a single image. When processing image
batches, you can use the
.Fl \-prefix Ar prefix
option, which constructs the name of each output file by prepending
the prefix
.Ar prefix
to the base name of the corresponding input file, and then
substituting
.Ql .pgm
with
.Ql .key
as before. Note that
.Ar prefix
is added verbatim to the name. If 
.Ar prefix 
is a directory, it should explicitly contain trailing file separator
.Ql / \.
.Pp
Descriptors can take a lot of disk space. This problem can be
alleviated by means of the
.Fl \-binary
option, which writes (only) the descriptors to a separate binary file.
The name of this file is obtained by removing the suffix
.Ql .key 
(if any) from the name of the keypoint output file, and then
appending the suffix
.Ql .desc 
to the result. Descriptors and are written in the same order of the
keypoints, with each component taking exactly one byte.
.Pp
Sometimes one wants to compute the descriptors of keypoints generated
externally. This can be done by the
.Fl \-keypoints Ar file.key
option, which reads the keypoints from
.Ar file.key .
This file has the same format of the keypoint files produced by
.Nm
(except that the descriptors are omitted). Keypoints are then written
along with their descriptors to the output file as normal.
Keypoints are processed by increasing scale, which means that
keypoints can appear in different order int the input and output
file.  Reordering can be prevented by means of the switch
.Fl \-stalbe\-oder .
If this option is used, it is reccomended to provide a list of
keypoints already ordered by scale as this may significantly improve
the computation speed. Note that
.Fl \-keypoints
is limited to process a single image per time.

.Sh EXAMPLES
.Bl -bullet
.It 
Read the image
.Ql image.pgm 
and write keypoints and descriptors to
.Ql image.key :
.Dl > sift image.pgm
.It
Same as above, but use null thresholds
.Dl > sift -t 0 -e 0 image.pgm
.It
Same as above, but do
.Em not
double the resolution of the image as part of the computation
of the Gaussian scale space:
.Dl > sift --first-octave 0 image.pgm
.It 
Read the image
.Ql image.pgm
and write the keypoints to
.Ql image.key
and the descriptor in binary format to
.Ql image.desc :
.Dl > sift -b image.pgm
.It 
Read the images
.Ql image1\&.pgm
andq
.Ql image2.pgm
and write the keypoints and descriptors to
.Ql /tmp/image1.key
and
.Ql /tmp/image2.key :
.Dl > sift -p /tmp/ image1.pgm image2.pgm
.It
Read the image
.Ql image.pgm
and the keypoints
.Ql image.key
and write keypoints and descriptors to
.Ql /tmp/image.key :
.Dl > sift -p /tmp/ -k image.key image.pgm
.It 
Same as above, but writes the descriptors in binary format to
.Ql /tmp/image.desc :
.Dl > sift -b -p /tmp/ -k image.key image.pgm
.El
.Sh SEE ALSO
.Xr convert 1 ,