樹莓派、樹莓派之學習、樹莓派之教育

W!o+ 的《小伶鼬工坊演義》︰神經網絡【FFT】二

什麼是事物的『特徵』呢？為什麼它的『提取方法』很重要？維基百科詞條這麼說︰

Feature extraction

In machine learning, pattern recognition and in image processing, feature extraction starts from an initial set of measured data and builds derived values (features) intended to be informative and non-redundant, facilitating the subsequent learning and generalization steps, and in some cases leading to better human interpretations. Feature extraction is related to dimensionality reduction.

When the input data to an algorithm is too large to be processed and it is suspected to be redundant (e.g. the same measurement in both feet and meters, or the repetitiveness of images presented as pixels), then it can be transformed into a reduced set of features (also named a features vector). This process is called feature selection. The selected features are expected to contain the relevant information from the input data, so that the desired task can be performed by using this reduced representation instead of the complete initial data.

───

 

假使考慮如何『定義』事物耶？也許『特徵』就是『界定性徵』，可以用來『區分』相異的東西！所以人們自然懂得『汪星人』不同於『喵星人』的也！！

於是乎好奇那『聲音』本有『調子』，可以用

Cepstrum

A cepstrum (/ˈkɛpstrəmˈˌˈsɛpstrəmˈ/) is the result of taking the Inverse Fourier transform (IFT) of the logarithm of the estimated spectrum of a signal. It may be pronounced in the two ways given, the second having the advantage of avoiding confusion with ‘kepstrum’ which also exists (see below). There is a complex cepstrum, a real cepstrum, a power cepstrum, and a phase cepstrum. The power cepstrum in particular finds applications in the analysis of human speech.

The name “cepstrum” was derived by reversing the first four letters of “spectrum”. Operations on cepstra are labelled quefrency analysis (aka quefrency alanysis[1]), liftering, or cepstral analysis.

Cepstrum_signal_analysis

Steps in forming cepstrum from time history

───

 

來探討。那麼『圖象』可有『調子』乎？！能否依樣畫葫蘆來研究的呢！？不管『笨鳥先飛』、『菜鳥忘飛』、『老鳥已飛』……… 科技史裡滿載『傻問題』之『大成就』矣！！？？何不就效法一下嘛？？！！

【還是用五】

Figure 5

 

>>> img = training_data[0][0].reshape(28,28)
>>> f_img = network.np.fft.rfft2(img)
>>> logp_img = 2*network.np.log(network.np.abs(f_img))
>>> plt.imshow(logp_img)
<matplotlib.image.AxesImage object at 0x51af290>
>>> plt.show()

 

Figure 5p

 

>>> ilogpf_img = network.np.fft.irfft2(logp_img)
>>> cf_img = network.np.abs(ilogpf_img)**2
>>> plt.imshow(cf_img)
<matplotlib.image.AxesImage object at 0x51d1050>
>>> plt.show()

 

Figure 5c

 

【依舊選零】

Figure 0

 

>>> img1 = training_data[1][0].reshape(28,28)
>>> f_img1 = network.np.fft.rfft2(img1)
>>> logp_img1 = 2*network.np.log(network.np.abs(f_img1))
>>> plt.imshow(logp_img1)
<matplotlib.image.AxesImage object at 0x5091e50>
>>> plt.show()

 

Figure 0p

 

>>> ilogpf_img1 = network.np.fft.irfft2(logp_img1)
>>> cf_img1 = network.np.abs(ilogpf_img1)**2
>>> plt.imshow(cf_img1)
<matplotlib.image.AxesImage object at 0x51b9b50>
>>> plt.show()
>>>

 

Figure 0c

 

【參考資料】

Discrete Fourier Transform (numpy.fft)

Standard FFTs

fft(a[, n, axis, norm]) Compute the one-dimensional discrete Fourier Transform.
ifft(a[, n, axis, norm]) Compute the one-dimensional inverse discrete Fourier Transform.
fft2(a[, s, axes, norm]) Compute the 2-dimensional discrete Fourier Transform This function computes the n-dimensional discrete Fourier Transform over any axes in an M-dimensional array by means of the Fast Fourier Transform (FFT).
ifft2(a[, s, axes, norm]) Compute the 2-dimensional inverse discrete Fourier Transform.
fftn(a[, s, axes, norm]) Compute the N-dimensional discrete Fourier Transform.
ifftn(a[, s, axes, norm]) Compute the N-dimensional inverse discrete Fourier Transform.

Real FFTs

rfft(a[, n, axis, norm]) Compute the one-dimensional discrete Fourier Transform for real input.
irfft(a[, n, axis, norm]) Compute the inverse of the n-point DFT for real input.
rfft2(a[, s, axes, norm]) Compute the 2-dimensional FFT of a real array.
irfft2(a[, s, axes, norm]) Compute the 2-dimensional inverse FFT of a real array.
rfftn(a[, s, axes, norm]) Compute the N-dimensional discrete Fourier Transform for real input.
irfftn(a[, s, axes, norm]) Compute the inverse of the N-dimensional FFT of real input.

Hermitian FFTs

hfft(a[, n, axis, norm]) Compute the FFT of a signal which has Hermitian symmetry (real spectrum).
ihfft(a[, n, axis, norm]) Compute the inverse FFT of a signal which has Hermitian symmetry.

Helper routines

fftfreq(n[, d]) Return the Discrete Fourier Transform sample frequencies.
rfftfreq(n[, d]) Return the Discrete Fourier Transform sample frequencies (for usage with rfft, irfft).
fftshift(x[, axes]) Shift the zero-frequency component to the center of the spectrum.
ifftshift(x[, axes]) The inverse of fftshift.

Background information

Fourier analysis is fundamentally a method for expressing a function as a sum of periodic components, and for recovering the function from those components. When both the function and its Fourier transform are replaced with discretized counterparts, it is called the discrete Fourier transform (DFT). The DFT has become a mainstay of numerical computing in part because of a very fast algorithm for computing it, called the Fast Fourier Transform (FFT), which was known to Gauss (1805) and was brought to light in its current form by Cooley and Tukey [CT]. Press et al. [NR] provide an accessible introduction to Fourier analysis and its applications.

Because the discrete Fourier transform separates its input into components that contribute at discrete frequencies, it has a great number of applications in digital signal processing, e.g., for filtering, and in this context the discretized input to the transform is customarily referred to as a signal, which exists in the time domain. The output is called a spectrum or transform and exists in the frequency domain.

Implementation details

There are many ways to define the DFT, varying in the sign of the exponent, normalization, etc. In this implementation, the DFT is defined as

A_k = \sum_{m=0}^{n-1} a_m \exp\left\{-2\pi i{mk \over n}\right\} \qquad k = 0,\ldots,n-1.

The DFT is in general defined for complex inputs and outputs, and a single-frequency component at linear frequency f is represented by a complex exponential a_m = \exp\{2\pi i\,f m\Delta t\}, where \Delta t is the sampling interval.

The values in the result follow so-called “standard” order: If A = fft(a, n), then A[0] contains the zero-frequency term (the mean of the signal), which is always purely real for real inputs. Then A[1:n/2] contains the positive-frequency terms, and A[n/2+1:] contains the negative-frequency terms, in order of decreasingly negative frequency. For an even number of input points, A[n/2] represents both positive and negative Nyquist frequency, and is also purely real for real input. For an odd number of input points, A[(n-1)/2] contains the largest positive frequency, while A[(n+1)/2] contains the largest negative frequency. The routine np.fft.fftfreq(n) returns an array giving the frequencies of corresponding elements in the output. The routine np.fft.fftshift(A) shifts transforms and their frequencies to put the zero-frequency components in the middle, and np.fft.ifftshift(A) undoes that shift.

When the input a is a time-domain signal and A = fft(a), np.abs(A) is its amplitude spectrum and np.abs(A)**2 is its power spectrum. The phase spectrum is obtained by np.angle(A).

The inverse DFT is defined as

a_m = \frac{1}{n}\sum_{k=0}^{n-1}A_k\exp\left\{2\pi i{mk\over n}\right\} \qquad m = 0,\ldots,n-1.

It differs from the forward transform by the sign of the exponential argument and the default normalization by 1/n.

───

 

竟然會看起來很像？？似乎又有點不一樣！！到底該說是『行』還是『不行』的呀？？？