From f58d5fde1ff3a95752210810cceb37398b185431 Mon Sep 17 00:00:00 2001
From: Guillaume HEUSCH <guillaume.heusch@idiap.ch>
Date: Thu, 28 Jun 2018 14:41:19 +0200
Subject: [PATCH] [extractor] fixed extractors for pulse-based PAD
---
bob/pad/face/extractor/FFTFeatures.py | 88 ----------------
bob/pad/face/extractor/FrequencySpectrum.py | 85 ---------------
bob/pad/face/extractor/LTSS.py | 110 +++++++++++++++-----
bob/pad/face/extractor/LiFeatures.py | 1 +
bob/pad/face/extractor/PPGSecure.py | 51 +++++----
bob/pad/face/extractor/__init__.py | 5 +-
6 files changed, 117 insertions(+), 223 deletions(-)
delete mode 100644 bob/pad/face/extractor/FFTFeatures.py
delete mode 100644 bob/pad/face/extractor/FrequencySpectrum.py
diff --git a/bob/pad/face/extractor/FFTFeatures.py b/bob/pad/face/extractor/FFTFeatures.py
deleted file mode 100644
index 48148832..00000000
--- a/bob/pad/face/extractor/FFTFeatures.py
+++ /dev/null
@@ -1,88 +0,0 @@
-#!/usr/bin/env python
-# encoding: utf-8
-
-import numpy
-
-from bob.bio.base.extractor import Extractor
-
-import logging
-logger = logging.getLogger("bob.pad.face")
-
-
-class FFTFeatures(Extractor, object):
- """
- Compute the Frequency Spectrum of the given signal.
-
- The computation is made using numpy's rfft routine
-
- **Parameters:**
-
- framerate: int
- The sampling frequency of the signal (i.e the framerate ...)
-
- nfft: int
- Number of points to compute the FFT
-
- debug: boolean
- Plot stuff
- """
- def __init__(self, framerate=25, nfft=256, concat=False, debug=False, **kwargs):
-
- super(FFTFeatures, self).__init__(**kwargs)
-
- self.framerate = framerate
- self.nfft = nfft
- self.concat = concat
- self.debug = debug
-
-
- def __call__(self, signal):
- """
- Compute the frequency spectrum for the given signal.
-
- **Parameters:**
-
- signal: numpy.array
- The signal
-
- **Returns:**
-
- freq: numpy.array
- the frequency spectrum
- """
- # sanity check
- if signal.ndim == 1:
- if numpy.isnan(numpy.sum(signal)):
- return
- if signal.ndim == 2 and (signal.shape[1] == 3):
- if numpy.isnan(numpy.sum(signal[:, 1])):
- return
-
- output_dim = int((self.nfft / 2) + 1)
-
- # get the frequencies
- f = numpy.fft.fftfreq(self.nfft) * self.framerate
-
- # we have a single pulse signal
- if signal.ndim == 1:
- fft = abs(numpy.fft.rfft(signal, n=self.nfft))
-
- # we have 3 pulse signal (Li's preprocessing)
- # in this case, return the signal corresponding to the green channel
- if signal.ndim == 2 and (signal.shape[1] == 3):
- ffts = numpy.zeros((3, output_dim))
- for i in range(3):
- ffts[i] = abs(numpy.fft.rfft(signal[:, i], n=self.nfft))
-
- if self.concat:
- fft = numpy.concatenate([ffts[0], ffts[1], ffts[2]])
- else:
- fft = ffts[1]
-
- if self.debug:
- from matplotlib import pyplot
- pyplot.plot(f, fft, 'k')
- pyplot.title('Power spectrum of the signal')
- pyplot.show()
-
- return fft
diff --git a/bob/pad/face/extractor/FrequencySpectrum.py b/bob/pad/face/extractor/FrequencySpectrum.py
deleted file mode 100644
index 059dd91a..00000000
--- a/bob/pad/face/extractor/FrequencySpectrum.py
+++ /dev/null
@@ -1,85 +0,0 @@
-#!/usr/bin/env python
-# encoding: utf-8
-
-import numpy
-
-from bob.bio.base.extractor import Extractor
-
-import logging
-logger = logging.getLogger("bob.pad.face")
-
-from scipy.signal import welch
-
-
-class FrequencySpectrum(Extractor, object):
- """
- Compute the Frequency Spectrum of the given signal.
-
- The computation is made using Welch's algorithm.
-
- **Parameters:**
-
- framerate: int
- The sampling frequency of the signal (i.e the framerate ...)
-
- nsegments: int
- Number of overlapping segments in Welch procedure
-
- nfft: int
- Number of points to compute the FFT
-
- debug: boolean
- Plot stuff
- """
- def __init__(self, framerate=25, nsegments=12, nfft=256, debug=False, **kwargs):
-
- super(FrequencySpectrum, self).__init__()
-
- self.framerate = framerate
- self.nsegments = nsegments
- self.nfft = nfft
- self.debug = debug
-
- def __call__(self, signal):
- """
- Compute the frequency spectrum for the given signal.
-
- **Parameters:**
-
- signal: numpy.array
- The signal
-
- **Returns:**
-
- freq: numpy.array
- the frequency spectrum
- """
- # sanity check
- if signal.ndim == 1:
- if numpy.isnan(numpy.sum(signal)):
- return
- if signal.ndim == 2 and (signal.shape[1] == 3):
- if numpy.isnan(numpy.sum(signal[:, 1])):
- return
-
- output_dim = int((self.nfft / 2) + 1)
-
- # we have a single pulse signal
- if signal.ndim == 1:
- f, psd = welch(signal, self.framerate, nperseg=self.nsegments, nfft=self.nfft)
-
- # we have 3 pulse signal (Li's preprocessing)
- # in this case, return the signal corresponding to the green channel
- if signal.ndim == 2 and (signal.shape[1] == 3):
- psds = numpy.zeros((3, output_dim))
- for i in range(3):
- f, psds[i] = welch(signal[:, i], self.framerate, nperseg=self.nsegments, nfft=self.nfft)
- psd = psds[1]
-
- if self.debug:
- from matplotlib import pyplot
- pyplot.semilogy(f, psd, 'k')
- pyplot.title('Power spectrum of the signal')
- pyplot.show()
-
- return psd
diff --git a/bob/pad/face/extractor/LTSS.py b/bob/pad/face/extractor/LTSS.py
index 5637b5d9..3b3555f1 100644
--- a/bob/pad/face/extractor/LTSS.py
+++ b/bob/pad/face/extractor/LTSS.py
@@ -5,15 +5,13 @@ import numpy
from bob.bio.base.extractor import Extractor
-import logging
-logger = logging.getLogger("bob.pad.face")
+from bob.core.log import setup
+logger = setup("bob.pad.face")
from scipy.fftpack import rfft
-
class LTSS(Extractor, object):
- """
- Compute Long-term spectral statistics of a pulse signal.
+ """Compute Long-term spectral statistics of a pulse signal.
The features are described in the following article:
@@ -29,38 +27,78 @@ class LTSS(Extractor, object):
year = 2017
}
- **Parameters:**
-
+ Attributes
+ ----------
framerate: int
The sampling frequency of the signal (i.e the framerate ...)
-
nfft: int
Number of points to compute the FFT
-
- debug: boolean
+ debug: bool
Plot stuff
+ concat: bool
+ Flag if you would like to concatenate features from the 3 color channels
+ time: int
+ The length of the signal to consider (in seconds)
+
"""
- def __init__(self, window_size=25, framerate=25, nfft=64, concat=False, debug=False, **kwargs):
+ def __init__(self, window_size=25, framerate=25, nfft=64, concat=False, debug=False, time=0, **kwargs):
+ """Init function
+
+ Parameters
+ ----------
+ window_size: int
+ The size of the window where FFT is computed
+ framerate: int
+ The sampling frequency of the signal (i.e the framerate ...)
+ nfft: int
+ Number of points to compute the FFT
+ concat: bool
+ Flag if you would like to concatenate features from the 3 color channels
+ debug: bool
+ Plot stuff
+ time: int
+ The length of the signal to consider (in seconds)
+ """
super(LTSS, self).__init__()
-
self.framerate = framerate
self.nfft = nfft
self.debug = debug
self.window_size = window_size
self.concat = concat
+ self.time = time
def _get_ltss(self, signal):
-
+ """Compute long term spectral statistics for a signal
+
+ Parameters
+ ----------
+ signal: numpy.ndarray
+ The signal
+
+ Returns
+ -------
+ ltss: numpy.ndarray
+ The spectral statistics of the signal.
+
+ """
+ window_stride = int(self.window_size / 2)
+
# log-magnitude of DFT coefficients
log_mags = []
- window_stride = int(self.window_size / 2)
# go through windows
for w in range(0, (signal.shape[0] - self.window_size), window_stride):
fft = rfft(signal[w:w+self.window_size], n=self.nfft)
mags = numpy.zeros(int(self.nfft/2), dtype=numpy.float64)
- mags[0] = abs(fft[0])
+
+ # XXX : bug was here (no clipping)
+ if abs(fft[0]) < 1:
+ mags[0] = 1
+ else:
+ mags[0] = abs(fft[0])
+ # XXX
+
index = 1
for i in range(1, (fft.shape[0]-1), 2):
mags[index] = numpy.sqrt(fft[i]**2 + fft[i+1]**2)
@@ -69,7 +107,6 @@ class LTSS(Extractor, object):
index += 1
log_mags.append(numpy.log(mags))
- # get the long term statistics
log_mags = numpy.array(log_mags)
mean = numpy.mean(log_mags, axis=0)
std = numpy.std(log_mags, axis=0)
@@ -78,18 +115,18 @@ class LTSS(Extractor, object):
def __call__(self, signal):
- """
- Computes the long-term spectral statistics for a given signal.
-
- **Parameters**
+ """Computes the long-term spectral statistics for given pulse signals.
- signal: numpy.array
+ Parameters
+ ----------
+ signal: numpy.ndarray
The signal
- **Returns:**
+ Returns
+ -------
+ feature: numpy.ndarray
+ the computed LTSS features
- feature: numpy.array
- the long-term spectral statistics feature vector
"""
# sanity check
if signal.ndim == 1:
@@ -100,9 +137,34 @@ class LTSS(Extractor, object):
if numpy.isnan(numpy.sum(signal[:, i])):
return
+ # truncate the signal according to time
+ if self.time > 0:
+ number_of_frames = self.time * self.framerate
+
+ # check that the truncated signal is not longer
+ # than the original one
+ if number_of_frames < signal.shape[0]:
+
+ if signal.ndim == 1:
+ signal = signal[:number_of_frames]
+ if signal.ndim == 2 and (signal.shape[1] == 3):
+ new_signal = numpy.zeros((number_of_frames, 3))
+ for i in range(signal.shape[1]):
+ new_signal[:, i] = signal[:number_of_frames, i]
+ signal = new_signal
+ else:
+ logger.warning("Sequence should be truncated to {}, but only contains {} => keeping original one".format(number_of_frames, signal.shape[0]))
+
+ # also, be sure that the window_size is not bigger that the signal
+ if self.window_size > int(signal.shape[0] / 2):
+ self.window_size = int(signal.shape[0] / 2)
+ logger.warning("Window size reduced to {}".format(self.window_size))
+
+ # we have a single pulse
if signal.ndim == 1:
feature = self._get_ltss(signal)
+ # pulse for the 3 color channels
if signal.ndim == 2 and (signal.shape[1] == 3):
if not self.concat:
diff --git a/bob/pad/face/extractor/LiFeatures.py b/bob/pad/face/extractor/LiFeatures.py
index 7e089b5c..9fbd4c48 100644
--- a/bob/pad/face/extractor/LiFeatures.py
+++ b/bob/pad/face/extractor/LiFeatures.py
@@ -68,6 +68,7 @@ class LiFeatures(Extractor, object):
-------
feature: numpy.ndarray
the computed features
+
"""
# sanity check
assert signal.ndim == 2 and signal.shape[1] == 3, "You should provide 3 pulse signals"
diff --git a/bob/pad/face/extractor/PPGSecure.py b/bob/pad/face/extractor/PPGSecure.py
index 09db66e7..bf4a8ddd 100644
--- a/bob/pad/face/extractor/PPGSecure.py
+++ b/bob/pad/face/extractor/PPGSecure.py
@@ -5,13 +5,12 @@ import numpy
from bob.bio.base.extractor import Extractor
-import logging
-logger = logging.getLogger("bob.pad.face")
+from bob.core.log import setup
+logger = setup("bob.pad.face")
class PPGSecure(Extractor, object):
- """
- This class extract the frequency features from pulse signals.
+ """Extract frequency spectra from pulse signals.
The feature are extracted according to what is described in
the following article:
@@ -30,39 +29,48 @@ class PPGSecure(Extractor, object):
year = 2017
}
- **Parameters:**
-
+ Attributes
+ ----------
framerate: int
The sampling frequency of the signal (i.e the framerate ...)
-
nfft: int
Number of points to compute the FFT
-
- debug: boolean
+ debug: bool
Plot stuff
+
"""
def __init__(self, framerate=25, nfft=32, debug=False, **kwargs):
-
- super(PPGSecure, self).__init__(**kwargs)
+ """Init function
+
+ Parameters
+ ----------
+ framerate: int
+ The sampling frequency of the signal (i.e the framerate ...)
+ nfft: int
+ Number of points to compute the FFT
+ debug: bool
+ Plot stuff
+ """
+ super(PPGSecure, self).__init__(**kwargs)
self.framerate = framerate
self.nfft = nfft
self.debug = debug
def __call__(self, signal):
- """
- Compute the frequency spectrum for the given signal.
-
- **Parameters:**
+ """Compute and concatenate frequency spectra for the given signals.
- signal: numpy.array
+ Parameters
+ ----------
+ signal: numpy.ndarray
The signal
- **Returns:**
-
- freq: numpy.array
- the frequency spectrum
+ Returns
+ -------
+ fft: numpy.ndarray
+ the computed FFT features
+
"""
# sanity check
assert signal.shape[1] == 5, "You should provide 5 pulses"
@@ -74,7 +82,7 @@ class PPGSecure(Extractor, object):
# get the frequencies
f = numpy.fft.fftfreq(self.nfft) * self.framerate
- # we have 5 pulse signal (Li's preprocessing)
+ # we have 5x3 pulse signals, in different regions across 3 channels
ffts = numpy.zeros((5, output_dim))
for i in range(5):
ffts[i] = abs(numpy.fft.rfft(signal[:, i], n=self.nfft))
@@ -94,5 +102,4 @@ class PPGSecure(Extractor, object):
logger.warn("Feature not extracted")
return
-
return fft
diff --git a/bob/pad/face/extractor/__init__.py b/bob/pad/face/extractor/__init__.py
index fabde901..c63151f5 100644
--- a/bob/pad/face/extractor/__init__.py
+++ b/bob/pad/face/extractor/__init__.py
@@ -5,10 +5,7 @@ from .VideoDataLoader import VideoDataLoader
from .VideoQualityMeasure import VideoQualityMeasure
from .FrameDiffFeatures import FrameDiffFeatures
-from .FrequencySpectrum import FrequencySpectrum
-from .FreqFeatures import FreqFeatures
-from .NormalizeLength import NormalizeLength
-from .FFTFeatures import FFTFeatures
+from .LiFeatures import LiFeatures
from .LTSS import LTSS
from .PPGSecure import PPGSecure
--
GitLab