Handayani Tjandrasa

Classification of non-proliferative diabetic retinopathy based on hard exudates using soft margin SVM

Diabetic retinopathy is a retinal disease caused by diabetes mellitus. Severity of diabetic retinopathy may lead to blindness. Therefore, early detection of diabetic retinopathy is very important. One of diabetic retinopathy symptoms is the existence of hard exudates. In this study, hard exudates in retinal fundus images are employed to classify the moderate and severe non-proliferative diabetic retinopathy. The hard exudates are segmented using mathematical morphology and the extracted features are classified by using soft margin SVM. The classification model achieves accuracy of 90.54% for 75 training data and 74 testing data of retinal images.

Classification of EEG Signals Using Single Channel Independent Component Analysis, Power Spectrum, and Linear Discriminant Analysis

Epilepsy is a neurological disorder of the brain that can generate epileptic seizures when abnormal excessive activity occurs in the brain. The seizure is marked by brief episodes of involuntary movement of the body, and sometimes followed by unconsciousness. In this study, the EEG classification system was performed to predict whether EEG signals belong to normal individuals, epileptic patients in seizure free or seizure condition. The EEG dataset contains 5 sets of 100 EEG segments which is referred to as set A to set E. The classification system consisted of three scenarios. One of the scenarios involved the methods of Single Channel Independent Component Analysis (SCICA), power spectrum, and a neural network. The results were compared to the results without implementing SCICA. The last experiment showed the effect of using Linear Discriminant Analysis (LDA) to reduce the features of power spectrum. The results gave the accuracies for 3, 4, and 5 classes. By applying SCICA, all the accuracies were improved significantly with the maximum accuracy of 94 % for 3 classes.

Soft Weighted Median Filter Method for Improved Image Segmentation with Noise

Soft Weighted Median Filter Method (SWMF) is one of the new methods for noise filtering in image processing. This method is used for two types of noise in images, there is fixed valued noise (FVN) and random valued noise (RVN). Fixed valued noise is a noise type with an unchanged value, it changes the pixel value of the image to the maximum and minimum values (0 and 255), while random valued noise is a noise type with a changed value. An example of fixed valued noise is salt & pepper noise, while for random valued noise can be exemplified as gaussian, poisson, speckle, and localvar noise. Based on previous research, SWMF method can be applied to all images with all kinds of noise (FVN and RVN) and able to reduce the noise well. This method has a higher PSNR value than other methods, especially for random valued noise types such as: gaussian, speckle, and localvar noise. In this study, we propose to examine the performance of the SWMF method further by comparing this method with other methods such as Median Filter, Mean Filter, Gaussian Filter, and Wiener Filter in an image segmentation process. The image segmentation process in this research is based on area detection using Top-Hat transform and Otsu thresholding and line detection using Sobel edge detection. The performance measurement process uses the calculation of sensitivity value, specificity, and accuracy on the image segmentation with the groundtruh image. The results show that Soft Weighted Median Filter method can improve the quality of image segmentation with the average accuracy of 95.70% by reducing fixed value noise and random valued noise in the images.

Alcoholism classification based on EEG data using Independent Component Analysis (ICA), Wavelet de-noising and Probabilistic Neural Network (PNN)

Alcoholism is a clinical symptom characterized by a tendency to drink more alcohol than planned or commonly called alcoholics. Alcoholics will suffer the damage in some parts of the body, including the brain. One way to detect alcoholics from the brain is to record the electrical activity of the brain through the scalp or called electroencephalography (EEG). EEG records are often disturbed by noise such as muscle movements, eye blinking and heartbeat. Therefore, this research suggests Independent Component Analysis (ICA), as noise removal, Stationary Wavelet Transform (SWT) as a feature extraction method and are classified into two classes, namely alcoholism and normal using Probabilistic Neural Network (PNN). In this research, the result obtained from the ICA noise removal, signal decomposition using Daubechies SWT at level 6 and Probabilistic Neural Network (PNN) is considered effective to extract features and classify the 64 channels alcoholism data. The data come from Neurodynamics Laboratory, State University of New York Health Center. The result of this research generate an accuracy of 85.00% from 100 random data trial using ICA, SWT decomposition level 6, Wavelet Daubechies type 4 and PNN deviation value of 0.6.

Feature extraction using combination of intrinsic mode functions and power spectrum for EEG signal classification

The measurement of brain electrical activity recorded as EEG signals finds most application in epilepsy. EEG waveforms carry information about the underlying neural system dynamics and show different features amongst epilepsy syndromes. In this research, empirical mode decomposition (EMD) and power spectrum were employed to extract the features from EEG dataset of healthy participants, and epilepsy patients with seizure and seizure free conditions. The recorded EEG signals are represented by 500 signal segments from 5 sets of different conditions. The sum of Intrinsic Mode Function (IMF) power spectrum components gave 10 features for 50 components, and 20 features for 25 components, which were used as the classification inputs for artificial neural networks and random forest classifiers. The classifications were carried out for 3, 4, and 5 classes. From the experiments, the highest average accuracy was obtained for 3 classes using 20 features of power spectrum from the sum of 6 IMFs. For use of 6 IMFs, the accuracies had the maximum values of 92.4%, 90.4%, and 78.6% for 3, 4, and 5 classes respectively. It also improved the accuracy significantly for 5 classes.

Feature extraction using statistical moments of wavelet transform for iris recognition

Iris is unique for each person, so that it can be used as one alternative solution for human identification. In this study, an iris recognition system is developed to automatically identify a person by using eye image data. Firstly, iris area of eye image is detected using Canny Edge Detection and Hough Transform methods. Secondly, texture feature of iris image is extracted using statistical moments of Wavelet Transform. Furthermore, the texture feature representation is recognized using Support Vector Machine classifier method. Experiment on CASIA eye image dataset gives good recognition rate, that is 93.5%.

Classification of EEG signals using Common Spatial Pattern-Principle Component Analysis and Interval Type-2 Fuzzy Logic System

Brain Computer Interface, defined as a direct communication pathway between human brain and computer, allows a system to get the intention of the brain via Electroencephalogram (EEG) signals. This mechanism thus does not involve the participation of motoric and muscular neurons. In recent progresses, things such as the variability of imagery activities and subject characteristics were found to be the main problems toward the development of reliable signal translation methods. In this paper, we propose an EEG signal translation system based on motoric imagery activities. The system includes band-pass filter and Common Spatial Pattern (CSP) for noise filtering and Principle Component Analysis (PCA) for feature extraction. Interval Type-2 Fuzzy Logic System is then used as the classifier for the produced features. The later identified classes, either 0 or 1, refer to the imagery cursor movement direction either upward or downward respectively. The training and testing data that used here are from BCI Competition II dataset 1a. The highest classification accuracy of the system was recorded at 85.2%.

New Lossless Compression Method using Cyclic Reversible Low Contrast Mapping (CRLCM)

Received Aug 03, 2016 Revised Nov 04, 2016 Accepted Nov 18, 2016 In general, the compression method is developed to reduce the redundancy of data. This study uses a different approach to embed some bits of datum in image data into other datum using a Reversible Low Contrast Mapping (RLCM) transformation. Besides using the RLCM for embedding, this method also applies the properties of RLCM to compress the datum before it is embedded. In its algorithm, the proposed method engages Queue and Recursive Indexing. The algorithm encodes the data in a cyclic manner. In contrast to RLCM, the proposed method is a coding method as Huffman coding. This research uses publicly available image data to examine the proposed method. For all testing images, the proposed method has higher compression ratio than the Huffman coding. Keyword:

Identification of oil palm plantation in IKONOS images using radially averaged power spectrum values

The use of satellite imagery for plantation management is helpful in monitoring the development of various parties including oil palm plantations. In a panchromatic IKONOS satellite imagery, oil palm plantations have unique characteristics that can be interpreted visually. This study tried to classify oil palm plantations from satellite imagery using texture characteristics with their spatial and frequency parameters. Spatial parameters are determined by calculating the first order features, while the second order texture variables are determined based on Gray Level Co-occurrence Matrix (GLCM), local feature, and Radially Average Power Spectrum Value (RAPSV). The classification accuracy of of this study reached 86%. An addition of average value of the power spectrum has increased the accuracy up to 28% compared to the usage of first order only.

A conceptual information of technology framework to support rice farming in timor leste

Agriculture is one of the main means of livelihood for the people of Timor Leste. Data shows that the Timor Leste annual rice consumption is 140 tonnes, while farmers only produce 60 tonnes per year [1]. One factor is that, the farmers still use traditional ways to tackle pests and diseases and when the weather forecast is not accurate. The condition of infrastructure, geographic, and communication services is a major obstacle for ministry of agriculture to have a monitoring, provide solutions, and control over agricultural issues in Timor Leste [2]. Therefore, it needs information technology to establish communication between farmers and the Ministry of Agriculture. This paper presents the multi-language framework concept namely Tetun, Indonesian, Portuguese, and English, because most of the people in Timor Leste still use the four languages for everyday communication. The aim of the framework concept is to describe a process of real time communication services with the farmers. The system acts as an expert to answer questions or to solve problems of farmers regarding pest and disease problems at any time, and hopefully it will increase agricultural productivity.