GLCM Texture Classification for EEG Spectrogram Image Mahfuzah Mustafa 1,2 , Mohd Nasir Taib 2,3 , Zunairah Hj. Murat 2,3 , Noor Hayatee Abdul Hamid 2,3 1 Faculty of Electrical & Electronics Engineering, Universiti Malaysia Pahang, Pahang, Malaysia mahfuzah@ump.edu.my 2 Faculty of Electrical Engineering, Universiti Teknologi MARA Malaysia, Selangor, Malaysia 3 Biomedical Research Laboratory for Human Potential, Universiti Teknologi MARA Malaysia, Shah Alam, Malaysia dr.nasir@ieee.org zunairahh@yahoo.com Abstract— Over the past century, time based and frequency based is used for analyzing Electroencephalography (EEG) signals. EEG is a scientific tool for measure signal from human brain. This paper proposes a time-frequency approach or spectrogram image processing technique for analyzing EEG signals. Gray Level Co-occurrence Matrix (GLCM) texture feature were extracted from spectrogram image and then Principal components analysis (PCA) was employed to reduce the feature dimension. The purpose of this paper is to classify EEG spectrogram image using k-nearest neighbor algorithm (kNN) classifier. The result shows classification rate was 70.83% for EEG spectrogram image. Keywords— EEG, spectrogram image, GLCM, texture feature, PCA, kNN I. INTRODUCTION The brain is the most complex organ in the human body. It performs various tasks such as control heart rate, hearing, speech and etc. The brain is made of billions cells called neurons. Neurons send a signal to control the movement of the whole human body. The signal or wave can be measured using EEG. However, this signal should be extracted to obtain information that can be used in research. EEG signal is analyzed by various methods, for example in time based, frequency based and time-frequency based. Usually, EEG raw signals are in time-based format. To analyze in frequency-based, usually the signals need to be transformed into Fourier Transform (FT). In this paper, EEG signals were analyzed based on time-frequency image processing technique or called spectrogram. The most often technique used to analyze signal in time-frequency based is Short Time Fourier Transform (STFT). The STFT is to perform a FT on the signal, then mapping the signal into a two-dimensional function of frequency and time. Many studies have been done on time-frequency based but mainly in signal processing area [1] but very few in image processing area. However, there is example of research using the spectrogram in time-frequency to classify heart abnormalities from electrocardiogram (ECG) [2]. The spectrogram was produced using STFT technique. They extracted the Euler number and height and width of pulses from the spectrogram image. For the purpose of classification, Back-propagation ANN was used and gives 100% accuracy in heart abnormalities. Once an image is obtained, various image processing tools, such as texture analysis, could be used. Most popular technique for the textural classification is GLCM. This technique commonly used to process texture of image various application such as satellite, wood and ultrasound. The GLCM is a tabulation of grey levels frequency occurring in an image. There is example of study use the GLCM in ECG [3]. The study is to detect sleep disorder breathing in human heart. They are using Fuzzy with result 79.29% accuracy in training and 75.88% accuracy in testing. The signal or wave can be measured using EEG. After GLCM process is done, texture features need to be extracted. The most popular texture feature extraction is proposed by Haralick [4] and Soh [5]. Haralick proposed 14 texture features to apply to the photomicrographs of sandstones, photographs of land-use and satellite photographs of land-use while Soh propose 10 texture features for satellite photographs of sea ice. It has been shown elsewhere, the PCA was used for data reduction and classification purposes. In [6], they choose the first three principal components which are contrast, dissimilarity and homogeneity, out of eight GLCM texture features. As a result, the first three components from PCA give better accuracy in classification than all eight GLCM texture features. In this paper, a study of dimension reduction of GLCM texture features is presented. The feature is extracted from EEG spectrogram image. The kNN is amongst the simplest classifier of machine learning algorithm. In kNN, an object is classified by a majority vote of its neighbors based on space. There are various types of space used, for example Euclidean, City block and Cosine but researchers often use the Euclidean distance. There are studies using Gray Level Co-occurrence Matrix (GLCM) texture feature to be fed to a kNN classifier [7, 8], in order to classify the images in various application. 2010 IEEE EMBS Conference on Biomedical Engineering & Sciences (IECBES 2010), Kuala Lumpur, Malaysia, 30th November - 2nd December 2010. 978-1-4244-7600-8/10/$26.00 ©2010 IEEE 373