Mehmet Cem Catalbas

Online signature recognition

In this paper, online signature recognition is examined by using K Nearest Neighborhood (KNN) method. The signatures are collected by an Android application which can extract the dynamic and spatial information of the signatures. In this frame, a signature database is consisting of a total of 120 signatures taken from 12 different person. The purpose of this paper, is to obtain high performance with a few training signatures. Also, the performance of signature recognition is investigated by different distance measurement methods in K Nearest Neighborhood.

Estimation of optimal locations for electric vehicle charging stations

Electric Vehicles (EV) have been commonly started to use due to some advantages such as less emission, lower noise pollution, maintenance requirement and power consumption. The number of charging stations have also increased based on rising the usage of EVs. Therefore, determination of optimal location for EV charge stations has a great importance for charging process. This localization is highly related with the range of EV and traffic density on areas. The distribution of charging stations is a basically optimization problem. For this reason, estimation of optimum locations for EV charging stations in Ankara, Turkey is realized by using data mining methods in this paper. Some parameters for determining of optimum locations which are the average number of EVs on the road and the average range are examined. Ankara road map is derived by using Mapbox Software obtained from the satellite via spectral clustering. Then, some of the image processing methods such as thresholding, erosion and dilation are used for eliminating clustering errors. Furthermore, optimal charging locations of EVs for Ankara are estimated by various clustering approaches such as spectral clustering and Gaussian Mixture Model (GMM) using a total number of charging stations. In conclusion, this paper is a novel study for Turkey which has not been worked in the literature and it can be easily applied to any region in the future works.

Modelling and estimation parameters of electronic differential system for an electric vehicle using radial basis neural network

This paper proposes modelling and estimation parameters of Electronic Differential System (EDS) for an Electric Vehicle (EV) with in-wheel motor using Radial Basis Neural Network (RBNN). In this study, EDS for front wheels is analysed instead of rear wheels which are commonly investigated in the literature. According to steering angle and speed of EV, the speeds of the front wheels are calculated by equations derived from Ackermann-Jeantand model using CoDeSys Software Package. The simulation of EDS is also realized by MATLAB/Simulink using the mathematical equations. Neural Network (NN) types including RBNN and Back-Propagation Feed-Forward Neural Network (BP-FFNN) are used for estimation the relationship between the steering angle and the speeds of front wheels. Besides, the different levels of noise are added to steering angle as sensor noise for realistic modelling. To conclude, the results estimated from types of NN are verified by CoDeSys and Simulink results. RBNN is convenient for estimation of EDS parameters due to robustness to different levels of sensor noise.

Optimal component selection for image segmentation via Parallel Analysis

In this paper, an image segmentation method is presented to analyze the clusters of Computed Tomography (CT) image. Target image is divided to small parts called as observation screens. Principal Component Analysis (PCA) is used for better representation of features about observation screens. The optimal number of component related with observation screen is determined by Horns Parallel Analysis (PA). Besides, Local Standard Deviation (LSD) which is a method for extracting meaningful sub-features is applied to whole image for successful segmentation. The effect of segmentation success rate is analyzed by selected features. Consequently, a novel algorithm is proposed for minimizing total computation time and error of dimension reduction significantly. It is seen that the results of the algorithm are approximately same as conventional segmentation algorithms.

Estimation of Trajectory and Location for Mobile Sound Source

In this paper, we present an approach to estimate mobile sound source trajectory. An artificially created sound source signal is used in this work. The main aim of this paper is to estimate the mobile object trajectory via sound processing methods. The performance of generalized cross correlation techniques is compared with that of noise reduction filters for the success of trajectory estimation. The azimuth angle between the sound source and receiver is calculated during the whole movement. The parameter of Interaural Time Difference (ITD) is utilized for determining azimuth angle. The success of estimated delay is compared with different types of Generalized Cross Correlation (GCC) algorithms. In this study, an approach for sound localization and trajectory estimation on 2D space is proposed. Besides, different types of pre-filter method are tried for removing the noise and signal smoothing of recorded sound signals. Some basic parameters of sound localization process are also explained. Moreover, the calculation error of average azimuth angle is compared with different GCC and pre-filtered methods. To conclude, it is observed that estimation of location and trajectory information of a mobile object from a stereo sound recording is realized successfully.

Morphological feature extraction with local histogram equalization

In this work, success of image feature extraction based on morphological is examined by various histogram equalization techniques. In comparison process, traditional histogram equalization, adaptive histogram equalization and local histogram equalization methods are used. In addition to that, effects on different structure element parameters is examined success of image feature extraction process.

Automatic determination of neovascularization area with adaptive histogram equalization

In this work, the determination of corneal neovascularization ratio is examined in the light of image processing techniques. The adaptive histogram equalization method for used to increase the success of retinal blood vessel areas on preprocess. Also, various morphological operations were performed on the cornea image to improve the performance of the blood vessel patterns determination.

Image classification via multi canonical correlation analysis

This work investigates the role of canonical correlations analysis in image classification problems. Canonical correlation analysis is proposed as an alternative feature selection and reduction method for generic image classification problems. This new method is studied via various image classification problems in comparison with principal components and kernel principal components analysis. Multiple canonical correlation analysis is proposed as a new feature selection and dimension reduction algorithm for image classification problems involving multiple classes. Classification performance and relationship between the extracted image attributes and classification performance are studied by using Caltech 101 dataset.

Country of origin estimation from composite faces using Kernel Principal Compenent Analysis

In this work, an algorithm is introduced that classifies test images into their originated countries using composite faces generated according to different countries. Also aim to increase success rate at implementation process using three color channel (R-G-B), color feature vector and local standard deviation matrix. Algorithm used Kernel Principal Component Analysis with gauss kernel structure for dimension reduction process. And optimal component number for dimension reduction process is determined via Horns parallel analysis method. At the end of process these obtained features are classified via Multi Support Vector Machines.

Super resolution using radial basis neural networks

The output of image size enlargement has important differences compared to the original sized image. In this study, an algorithm which intends to minimize the loss due to these differences, is presented. This minimization process is provided by radial bases neural networks (RBNN). In order to achieve better performance the RBNN activation function radius criteria is chosen adaptively throughout the work. It is observed that this new proposed method achieves better performance than that of methods in the literature. With the use of this method, it is foreseen that human made mistakes in disease diagnosis like computer tomography, inwhich small details are important, will be reduced.