Brain oscillations are very powerful descriptors of both physiological and pathological brain states. In general, EEG signals consist of complex mixtures of components whose characterization provides reliable information about the neuronal activity. This study is inspired to the {\em consensus matching pursuit} (CMP) representation and proposes an effective method for the detection and modeling of interictal prototypical signal patterns in temporal lobe epilepsy. CMP allows accounting for inter-trial variability in temporal jitter, frequency and number of oscillations. In this work, we propose to generalize the approach and exploit the resulting spike representation for automatic interictal spike detection. Performance was evaluated on both synthetic and real high density EEG signals. Results show high sensitivity and specificty in spike detection as well as an accurate separation in the transient and oscillation components.
