Epilepsy is a socially-stigmatizing chronic neurological condition. Limited availability of seizure Electroencephalogram (EEG) data makes the application of machine learning techniques for epileptic seizure detection very challenging. In this work, an efficient algorithmic procedure is proposed to facilitate the learning and classification of epileptic seizures from imbalanced EEG data. We designed an end-to-end architecture by combining local binary pattern with Siamese convolutional neural network. We used local binary pattern due to its capability to capture distinguishable morphological characteristics in the EEG signal. Siamese convolutional neural network was used since it can learn a similarity metric using an extremely small number of training samples for seizure episodes. With availability of a very small amount of training (seizure) samples, the effectiveness of the proposed method was verified by comparing the Siamese convolutional neural network with a baseline convolutional neural network. The proposed architecture outperforms the baseline model and achieves an average of 11.66% increase in F1-measure.