Figure 1.
The selected electrode locations of the International 10–20 system (29 EEG recording electrodes (black circles), one ground and one reference electrode (red circles) used in this paper).
Figure 2.
Examples of the face stimuli obtained by the 3dMD face capture system with the same illumination conditions.
One self-face image and a number of non-self-face images are depicted here.
Figure 3.
The different conditions of the paradigm.
(a) The classical row-column (RC) paradigm, (b) The proposed random set presentation (RASP) paradigm, (c) RASP paradigm with flashing self-face in one row of the virtual matrix, (d) RASP paradigm with flashing non-self-face in one column of the virtual matrix. Both RASP and RASP-F stimuli were shown semi-transparently to the participants such that the characters were still visible. However, this is not shown here for illustration purposes.
Figure 4.
Classification accuracy curves of each subject for three conditions ('RC', 'RASP', and 'RASP-F') using one to ten sequences.
Also, on the bottom right the averaged classification accuracy and ITR on all subjects are plotted.
Table 1.
F-values and significance of the repeated measures ANOVA.
Table 2.
Single-trial classification accuracy [%], based on 8-fold cross validation for each paradigm and subject.
Figure 5.
Topographic plots of grand average ERP waveform derived from the target and non-target stimuli for all 15 participants at 29 electrode channels ((a) RC, (b) RASP, (c) RASP-F).
The scales of x- and y-axes for each channel are the same.
Figure 6.
Grand average ERPs and scalp topographies for the three conditions RC, RASP, and RASP-F.
Top row: ERPs for targets and nontargets at two selected electrodes Cz and PO7. The two shaded areas in each ERP plot mark the intervals for which scalp maps are shown underneath. Center: The first and second row of scalp plots indicate the ERP responses to the target and nontarget classes. Bottom row: Temporal distribution based on sgn r at two selected electrodes Cz and PO7. The P300 component show a higher discriminability for RASP-F as compared to the two other spellers at the central and parieto-occipital sites.
Figure 7.
Grand average ERPs and scalp topographies for the self-face and non-self-face stimuli of the RASP-F condition.
Top row: ERPs for targets and nontargets at two selected electrodes Cz and PO7. The three shaded areas extract 3 discriminative intervals. Scalp maps are shown underneath using these intervals. Center: The first and second row of scalp plots indicate the ERP responses to the target and nontarget classes. Bottom row: Temporal distribution based on sgn r at two selected electrodes Cz and PO7. ERPs resulting from the self-face stimulus show stronger responses as those of the non-self face stimulus.
Figure 8.
values, comparing target stimuli for the three considered paradigms (A,B,C) as well as comparing the two types of target stimuli in RASP-F: non-self-face and self-face stimuli (D). Time courses of
values are given below for two EEG channels (namely 'Cz' and 'PO7').
Table 3.
Examines differences of ERP components with respect to target stimuli.
Figure 9.
Performance comparison across spellers as number of sequences is increased.
Blue circles used one repetition and red stars three repetitions. (a) Error distributions for the RC (left), RASP (center), and RASP-F (right). All target items have been centered in each matrix. The number in a black centered cell corresponds the number of correct selections and numbers in other cells represents the number of error corrected selections occurring in each cell relative to the target location for each speller. (b) Scatter plot comparing classification accuracies and significance values of various combinations of the three conditions. Each circle represents the classification accuracy of one subject.
Figure 10.
Compares the classification accuracy (y-axis) when increasing the number of non-targets preceding a target stimulus for the three conditions (x-axis).
The value '0' of x-axis presents 'double flashed target'.