Automated Detection of Epileptic Seizure Using Histogram of Oriented Gradients for Analysing Time Frequency Images of EEG Signals

Abstract

This paper presents a computer aided methodology for automated detection of epileptic seizure activities using the histogram of oriented gradient (HOG) descriptor for analysing the time-frequency(t-f) representation (image) of EEG signals. The proposed approach for classifying epileptic signals is based on transforming the EEG signal into a t-f image and the features related to t-f structures and shapes are extracted using HOG descriptor. The obtained feature is fed into SVM classifier with Gaussian kernel function for classification process. Further, the computationally efficient techniques based on local binary patterns(LBP) namely center symmetric LBP(CS-LBP) and local binary count(LBC) are also proposed to analyse the t-f representation of EEG signals. The histogram features extracted from CS-LBP and LBC are fed into SVM classifier. The performances of the proposed techniques are compared in terms of computational performance and detection rate of the classifier. Experimental evaluation on publicly available EEG dataset suggests that features extracted using HOG technique are more powerful in classifying epileptic and non-epileptic signals with an accuracy rate of 100%. Further, the obtained result suggests that proposed HOG technique outperforms the existing technique in the literature and attains better classification accuracy with an improved computational performance.