**3. Feature extraction and classification algorithms**

To increase the performance of the HCIs, algorithmic studies related to feature extraction and classification were realized. Motor imagery based BCI, the feature extraction, was performed with an adaptive autoregressive model and the classifier used was an adaptive quadratic discriminant analysis (Vidaurre et al., 2006). A new algorithm for single-trial online classification of imagery left and right hand movements was developed. This algorithm is based on time-frequency information derived from filtering EEG wideband raw data with causal Morlet wavelets, which are adapted to individual EEG spectra (Lemm et al., 2004). For motor imagery EEG, a new EEG recognition algorithm which combined the discrete wavelet transform with the backpropagation neural network was developed (Ming-Ai et al., 2009). According to the results, performance of motor imagery based BCI using a single recording session of EEG or ECoG signals for each subject, is not sufficient. It was relatively easy to obtain classifiable signals quickly from most of the non-paralyzed subjects. However, it was proved that it is impossible to classify the signals obtained from the paralyzed patients by the same methods (Hill, et al., 2006). To detect the ERPs, EEG

The behavior of active motor units identified via analysis of EMG signals recorded from the first dorsal interosseous muscle using a quadrifilar needle electrode is investigated. According to this study, the motor unit action potential waveforms recorded from patients were more complex than those recorded from control subjects as often observed in motor neuron diseases (Kasi et al., 2009). An eating assistant robot used to assist in eating independence was developed. This assistant robot is useful for people with severe disabilities. A spoon and a camera are attached on the tip of the robotic arm (Takahashi et al., 2001). Additionally, detecting the stress level of the computer user could possibly develop the computers' ability to respond intelligently and help calm negative emotional

To increase the performance of the HCIs, algorithmic studies related to feature extraction and classification were realized. Motor imagery based BCI, the feature extraction, was performed with an adaptive autoregressive model and the classifier used was an adaptive quadratic discriminant analysis (Vidaurre et al., 2006). A new algorithm for single-trial online classification of imagery left and right hand movements was developed. This algorithm is based on time-frequency information derived from filtering EEG wideband raw data with causal Morlet wavelets, which are adapted to individual EEG spectra (Lemm et al., 2004). For motor imagery EEG, a new EEG recognition algorithm which combined the discrete wavelet transform with the backpropagation neural network was developed (Ming-Ai et al., 2009). According to the results, performance of motor imagery based BCI using a single recording session of EEG or ECoG signals for each subject, is not sufficient. It was relatively easy to obtain classifiable signals quickly from most of the non-paralyzed subjects. However, it was proved that it is impossible to classify the signals obtained from the paralyzed patients by the same methods (Hill, et al., 2006). To detect the ERPs, EEG

Fig. 10. The block diagram of a NIR system.

**2.4 Other approaches** 

states of the user during HCI.

**3. Feature extraction and classification algorithms** 

recordings are transformed into a Haar-wavelet series (Kawakami et al., 1996) and variational Kalman filtering (Sykacek et al., 2004) for adaptive classification in the BCI system was used. The later algorithm translates EEG segments adaptively into probabilities of cognitive states. It allows for nonstationarities in the joint process over cognitive states and generated EEG which may occur during a consecutive number of trials. The wavelet features are used to determine the characteristic of eye movement waveform (Daud & Sudirman, 2010).

A new two-stage approach to extract the µ rhythm component was developed. The first stage uses second-order blind identification with stationary wavelet transform to automatically remove the artifacts. In the second stage, second-order blind identification is applied again to find the µ rhythm component. In this method artifact removal enhances the extraction of the µ rhythm component (Ng & Raveendran, 2009). For classification of motor execution signals, fractal approach provides promising results (Usakli, 2010). An EEG based BCI for users to control a cursor on a computer display is one of the common study area. The developed system uses an adaptive algorithm, based on kernel partial least squares classification, to associate patterns in multichannel EEG frequency spectra with cursor controls (Trejo et al., 2006). For the BCI related classification review can be found in (Lotte et al., 2007).
