Brain–computer interfaces (BCIs) enable communication and device control without muscle movement and offer important benefits for people with severe motor impairments. This thesis develops a more efficient EEG‑based BCI that detects event‑related potentials (ERPs) from single trials, improving throughput for visual image search. Using cross‑session training, a hybrid generative–discriminative classifier, and computationally efficient learning methods, the system achieves strong ERP detection without performance loss. The thesis also identifies a neural correlate of visual perception by linking ERP responses to task and target difficulty, suggesting that the brain dynamically increases cognitive resource allocation during challenging visual‑recognition tasks.
