Introduction
-
The MKDE employing the KDE combined with the K-means and the multiscale strategy can explore the fine-grained information hidden in the original data and better estimate the marginal density function.
-
The RCM estimated by correlation analysis based on the remedied data obtained by the dictionary learning may contain less pseudo-dependence, which could be closer to the real correlation matrix parameter of the copula density function.
-
The proposed EEG-based copula model is used to assess visual discomfort induced by the stereoscopic display. Compared with the time-domain EEG features extracted by the state-of-the-art method, the features extracted by our proposed model present better discrimination to visual discomfort.
Primitive Copula Model
Improved EEG-Based Copula Model
Mixture Kernel Density Estimation
Remedied Correlation Matrix
Evaluation of the Improved EEG-based Copula Model
Simulation Experiment
Behaviour Experiment
EEG Experiment
Framework of Assessing Visual Discomfort
Experimental Results
Simulation Experiments on MKDE
Simulation Experiments on RCM
Experiments on Assessing Visual Discomfort
Features | Linear concatenation |
---|---|
K_O | Smoothing parameter from the KDE and the correlation matrix parameter of the copula function obtained by the original data |
K_R | Smoothing parameter from the KDE and the RCM |
1C_M_R | Optimal smoothing parameter from one cluster with the largest numbers of data and the RCM |
1C_3P_M_R | First three smoothing parameters from one cluster with the largest numbers of data and the RCM |
2C_M_R | Optimal smoothing parameters from two clusters with the first two largest numbers of data and the RCM |
2C_2P_M_R | First two optimal smoothing parameters from two clusters with the first two largest numbers of data and the RCM |
2C_3P_M_R | Three smoothing parameters from two clusters with the first two largest numbers of data and the RCM |
3C_M_R | Optimal smoothing parameters from three clusters of data and the RCM |
3C_3P_M_R | All the smoothing parameters from three clusters of data and the RCM |
Method of feature extraction | Features | Accuracy (mean ± SD) |
---|---|---|
AR model | coefficients (rough EEG features) | 0.4856 ± 0.0018 |
median of coefficients | 0.5060 ± 0.0075 | |
mean of coefficients | 0.4947 ± 0.0065 | |
standard deviation of coefficients | 0.5049 ± 0.0082 | |
kurtosis of coefficients | 0.5098 ± 0.0055 | |
skewness of coefficients | 0.5073 ± 0.0055 | |
combined statistics of coefficients | 0.5059 ± 0.0024 | |
Entropy measures | sample entropy | 0.4873 ± 0.0863 |
multiscale entropy (state-of-the-art method) | 0.4562 ± 0.0945 | |
Proposed model | 2C_M_R | 0.8163 ± 0.0108 |