China
Africa
Explore chapters and articles related to this topic
A Review on Face and Gait Recognition: System, Data, and Algorithms
Published in Alexander D. Poularikas, Stergios Stergiopoulos, Advanced Signal Processing, 2017
Haiping Lu, Jie Wang, Konstantinos N. Plataniotis
The algorithms reviewed so far all take vectorial input. However, gray-level face images () and binary gait silhouette sequences (row ´ column row ´ column ´ time) are naturally multidimensional objects, which are formally called tensor objects. Therefore, the linear and nonlinear algorithms above need to reshape these tensors into vectors in a very high-dimensional space, which not only results in high computation and memory demand, but also breaks the natural structure and correlation in the original data. This motivated the recent development of the multilinear subspace learning algorithms [78–84], which extract features directly from the tensorial representation rather than the vectorized representation, and it is believed that more compact and useful features can be obtained this way.
The internet of things for smart manufacturing: A review
Published in IISE Transactions, 2019
Hui Yang, Soundar Kumara, Satish T.S. Bukkapatnam, Fugee Tsung
Pattern recognition and feature extraction: Data representation and visualization help transform the raw data to alternative domains, e.g., frequency domain, wavelet domain, and state-space domain. The next step is to learn and recognize hidden patterns using pattern recognition methods such as principal component analysis, data clustering, factor analysis, multilinear subspace learning, and Bayesian networks. Further, feature extraction focuses on the quantification of salient patterns as features for system informatics and control. For examples, Bukkapatnam et al. (1999a, 1999b) proposed the wavelet analysis of acoustic emission signals for feature representation in metal cutting Koh et al. (1995) integrated engineering knowledge with the Haar transformation for tonnage signal analysis and fault detection in stamping processes Jin and Shi (1999, 2000) developed feature-preserving data compression of stamping tonnage signals using wavelets, and further decomposed press tonnage signals to obtain individual station signals in transfer or progressive die processes. Ding, Zeng et al. (2006) proposed the integration of data-reduction with data-separation tasks for process monitoring and statistical control of waveform signals. Yang et al. (2007) also proposed an adaptive wavelet method to represent nonlinear dynamic signals for feature extraction in the state space. Bukkapatnam et al. (2002) and Bukkapatnam et al. (2009) developed local Markov models to predict system dynamics and future evolution in the state space. Yang and Chen (2014) and Chen and Yang (2016b) also developed a new heterogeneous recurrence approach to monitor and control nonlinear stochastic processes. Heterogeneous recurrence analysis was successfully implemented for both sleep apnea monitoring (Cheng et al., 2016) and the identification of dynamic transitions in ultraprecision machining processes (Kan et al., 2016).