1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
|
.TH WZIP 1 "24 December 1997"
.\" NAME should be all caps, SECTION should be 1-8, maybe w/ subsection
.\" other parms are allowed: see man(7), man(1)
.SH NAME
wzip \- lossy data compression and denoising
.SH SYNOPSIS
.B wzip
.I "[ -c | -d | -dn | -hdn ] num sf"
.SH "DESCRIPTION"
This manual page documents the
.BR wzip
command.
.PP
.B wzip
is a program that can be used for LOSSY data compression and denoising.
It reads from STDIN and writes to STDOUT. In compression mode the input
is a sequence of ascii floating-point values.
.B num
is the number of these
data values. The output is a sequence of small integers, most of
them zero in typical application. This is ready for effective compression
with a standard loss-less compression program like gzip.
.PP
The program can also be used for denoising. In this case both input
and output are sequences of ascii floating-point values.
.PP
The scale factor
.B sf
determines the strength of compression or denoising.
A higher scale factor means heavier compression and stronger denoising.
Four times the standard deviation of the noise content is a good start.
Otherwise 5 percent of the overall signal amplitude might be used as a
first estimation of a suitable scale factor.
.PP
If the noise content of the input data is strongly non-Gaussian-distributed,
like Poisson noise. The input data should be transformed to approximate
Gaussian-distributed noise. If the input values are Poisson-distributed,
that means for example raw counts per channel in EDX or XPD, they can
be transformed to approximate Gaussian-distributed noise by transformation
of each data point with y:=2.0*sqrt(x+0.25109). Back transformation
is done with y:=(x/2)^2. The summand 0.25109 compensates for the bias
caused by the asymmetry of the Poisson-distribution.
.PP
Invoking the program without any options writes examples of the use of
the program to STDERR.
.SH OPTIONS
There must be given exactly one option.
.TP
.B \-c
Compression, reads
.B num
ascii floating-point values from STDIN and writes a sequence of integers with
high redundancy to STDOUT.
.TP
.B \-d
Decompression, reads from STDIN and writes a sequence of
.B num
ascii floating-point values to STDOUT. These are more or less similar to
the original data.
.TP
.B \-dn
Denoising, reads
.B num
ascii floating-point values from STDIN and writes a sequence of
.B num
ascii floating-point values to STDOUT. These are more or less similar to
the original data.
.TP
.B \-hdn
Denoising with hard thresholding instead of wavelet shrinkage. Single
untouched noise peaks may be visible with this mode. On the other hand,
there is much less impact on the signal slope.
.SH "SEE ALSO"
Donoho, D.L.; Johnstone, I.M.: Adapting to unknown smoothness via wavelet
shrinkage, technical report 425, Department of Statistics, Stanford
University, Stanford, June 1993,
ftp://playfair.stanford.edu/pub/donoho/ausws.ps.Z
.PP
Franzen, A.: Compression of process data with a wavelet method, steel
res. 69 (1998), No. 1, pp. 28/30
.PP
Franzen, A.: Non-linear denoising with wavelet transformation, Z. Metallkd.
89 (1998), No. 4, pp. 297/302
.SH AUTHOR
This manual page was written by Andreas Franzen <anfra@debian.org>,
for the Debian GNU/Linux system (but may be used by others).
.P
Copyright (C) 1997 Andreas Franzen, placed under the GNU General Public
License, see the file copyright for details.
|
