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
90
91
92
93
94
95
96
|
## SIGTOOLS module
##
#Copyright (c) 2002 Travis Oliphant all rights reserved
#Oliphant.Travis@altavista.net
#Permission to use, modify, and distribute this software is given under the
#terms of the SciPy (BSD style) license. See LICENSE.txt that came with
#this distribution for specifics.
I hope that this packages grows into a set of widely useful 1-D, 2-D,
3-D, etc. filtering routines in this collection. To start it off I
have created C-code to do fast, general filtering of data. Most
additional code to be added should probably be written in Python
unless a specific need for speed is needed. If you have good routines
to add to this collection (in Python or in C) please send them to me.
So far this library has 4 C-code routines:
(1) a general purpose N-D correlation. This is almost like N-D
convolution but performs no 180 degree flipping of either
input to the routine. It takes two N-D arrays and returns
the correlation of the two arrays. It is intended for use in
filtering large arrays with small kernels for blurring,
edge-detection, etc. Zero-padding is used to handle
edges.
(2) a general purpose N-D order-filter routine
An order-filter takes a data array and a domain kernel with
the same number of dimensions which has zero and non-zero
entries. Each output is the order_th element of a sorted
list composed of the elements in the data array where the
kernel window (centered on the output point) is non-zero.
Zero-padding is used to handle the edges.
(3) a 1-D linear filter.
linear_filter is a replacement for MATLAB's filter function.
It takes two 1-D arrays representing the numerator and
denominator of the z-transform of a rational linear
system. It then performs the filtering along the given
axis of an N-D array. Access to initial conditions on the
Direct Form II Transposed system are provided.
(4) a remez exchange algorithm.
remez is Python wrapping for code from Erik Kvaleberg (itself a
conversion to C of original code by Mcclellan) which
implements the Parks-McClellan alogrithm for FIR optimal
filter design.
In Python several more routines are defined:
convolveND (performs 180 degree flipping of all axes)
wiener (implements a wiener filter in N-D)
medfilt (implements an N-D median filter)
All of these methods operate on arbitrary Python sequence objects (even
NumPy arrays of objects).
Due to popular request I have also included the numpyio module as part of
this package. It compiles along with the sigtools code.
----------------------------------------
INSTALLATION
----------------------------------------
To install, copy Makefile.pre.in from your python installation directory
(mine is /usr/lib/python1.5/config/Makefile.pre.in) to this directory.
Type
make -f Makefile.pre.in boot
make
Copy the file to somewhere under $PYTHON/site-packages or somewhere
else on your path.
Please send me feedback, questions, and especially contributions.
CHANGELOG
============
Feb 5, 1999 First release just contains convolveND
Feb 8, 1999 Second release (0.20) added order_filterND
Feb 23, 1999 Third release (0.40) added linear_filter and remez
routines.
May 1, 1999 Release 0.5.1 introduces the signaltools.py module which
has some filtering routines. Fixed a small bug with complex
filters that was preventing them from functioning.
July 10, 1999 Release 0.5.2 is incorporated now with multipack. Changed
the remez.c code to that from Erik Kvaleberg so that it
could be incorporated into and LGPL package. Placed remez
code directly in sigtoolsmodule.c
Travis Oliphant
oliphant.travis@ieee.org
|
