China 
Africa 
Explore chapters and articles related to this topic
Wavelets and Wavelet Transform
Published in Sing-Tze Bow, Pattern Recognition and Image Preprocessing, 2002
A signal usually consists of low-frequency and high-frequency contents. The low-frequency content is usually the most important part of the signal, as it gives the signal its identity. The high-frequency content imparts flavor or nuance. There are two technical terms conventionally used in the wavelet analysis, namely, approximations A and details D. Approximations refers to the high-scale-factor, low-frequency components of the signal, which can be matched with the stretched wavelets, while details refers to the low-scale-factor, high-frequency components which are to be matched by the compressed wavelets. These two component parts of the signal can be separately extracted through a filter bank. A filter bank is a set of filters used to separate an input signal into frequency bands for analysis. For our case two filters are usually chosen for the bank, the high-pass and the low-pass filter. The high-scale-factor, low-frequency components of the signal can pass through the low-pass filter, while the high-frequency components of the signal (i.e., the low-scale-factor components) are singled out at the output of the high-pass filter.
1
Published in Rajarshi Gupta, Dwaipayan Biswas, Health Monitoring Systems, 2019
Simone Benatti, Victor Kartsch, Fabio Montagna, Elisabetta Farella, Velu P. Kumaravel
The wavelet transform can be considered a bandpass filter and the series of scaled wavelet a filter bank. Using a finite number of values for the decomposition, we extract information in time-frequency domain by the recursive filtering of the given signal. Figure 6.11 shows the sequences of high and low pass filtering sequence performed to obtain the DWT coefficients. For a DWT decomposition of level n, the coefficients of the low pass filters are named detail coefficients (Dn) while the coefficients of the high pass filters are named approximation coefficients (An).
Video Compression
Published in Keshab K. Parhi, Takao Nishitani, Digital Signal Processing for Multimedia Systems, 2018
A digital filter bank is a collection of filters with a common input (referred to as the analysis filter bank) or a common output (referred to as the synthesis filter bank). Filter banks are generally used for subband coding, where a single signal x(n) is split into m subband signals in the analysis filter bank; in the synthesis filter bank, m input subband signals are combined to reconstruct the signal y(n).
Electromagnetic field and artificial intelligence based fault detection and classification system for the transmission lines in smart grid
Published in Energy Sources, Part A: Recovery, Utilization, and Environmental Effects, 2021
Chetan Khadse, Abhijeet A. Patharkar, Bharat S. Chaudhari
The preprocessing or feature extraction process of magnetic field signals under different fault conditions are done with the FIR filter. The dyadic analysis filter bank is used for creating filter bank. The input to the filter bank is treated as frames and given to the filters like discrete Meyer, biorthogonal, Daubechies, Haar, and other wavelet families. Daubechies four wavelets with four levels are chosen for the feature extraction. As large number of cases are considered in the simulation, massive number of values of Hx and Hy are generated for each case. All these magnetic field values are given to the dyadic analysis filter bank. The and are decomposed into a database of subbands with slower sample rates and smaller bandwidths. The tree structure is set to the symmetric so that filter bank act as discrete wavelet transform which gives subbands (Fliege 1994). These subbands are of the size [length of input/16] as four-level Daubechies wavelet is chosen. These generated dataset are the extracted features which contain information about energies of input signal. The features are used as input signal to train the neural network.
Review of Time–Frequency Masking Approach for Improving Speech Intelligibility in Noise
Published in IETE Technical Review, 2021
For the binary classification or regression, acoustic features are required to be extracted for each T–F unit. Before extracting features, in order to decompose sound mixture into frequency bands, the input signal is fed into a filterbank which consists of human auditory modeling filters, e.g. gammatone filters [28] or mel [24,25] or Bark scale filters. The output of each auditory filter is then divided into 20-ms segments with 50% overlap yielding T–F units. In the speech and speaker recognition literature, several acoustic features have been considered [41–44]. The following outlines several acoustic features which have been explored for T-F mask estimation.