Oğuz Findik

A directed artificial bee colony algorithm

A new method based on artificial bee colony (ABC) algorithm is proposed in this study.The improvement is based on direction information produced for artificial bees.Performance of the proposed method has been examined on numeric functions.The experimental results show that proposed approach is more effective than some classical variants of ABC algorithm. Artificial bee colony (ABC) algorithm has been introduced for solving numerical optimization problems, inspired collective behavior of honey bee colonies. ABC algorithm has three phases named as employed bee, onlooker bee and scout bee. In the model of ABC, only one design parameter of the optimization problem is updated by the artificial bees at the ABC phases by using interaction in the bees. This updating has caused the slow convergence to global or near global optimum for the algorithm. In order to accelerate convergence of the method, using a control parameter (modification rate-MR) has been proposed for ABC but this approach is based on updating more design parameters than one. In this study, we added directional information to ABC algorithms, instead of updating more design parameters than one. The performance of proposed approach was examined on well-known nine numerical benchmark functions and obtained results are compared with basic ABC and ABCs with MR. The experimental results show that the proposed approach is very effective method for solving numeric benchmark functions and successful in terms of solution quality, robustness and convergence to global optimum.

A comparison of feature selection models utilizing binary particle swarm optimization and genetic algorithm in determining coronary artery disease using support vector machine

The aim of this study is to search the efficiency of binary particle swarm optimization (BPSO) and genetic algorithm (GA) techniques as feature selection models on determination of coronary artery disease (CAD) existence based upon exercise stress testing (EST) data. Also, increasing the classification performance of the classifier is another aim. The dataset having 23 features was obtained from patients who had performed EST and coronary angiography. Support vector machine (SVM) with k-fold cross-validation method is used as the classifier system of CAD existence in both BPSO and GA feature selection techniques. Classification results of feature selection technique using BPSO and GA are compared with each other and also with the results of the whole features using simple SVM model. The results show that feature selection technique using BPSO is more successful than feature selection technique using GA on determining CAD. Also with the new dataset composed by feature selection technique using BPSO, this study reached more accurate values of success on CAD existence research with more little complexity of classifier system and more little classification time compared with whole features used SVM.

Effects of principle component analysis on assessment of coronary artery diseases using support vector machine

Artificial intelligence techniques are being effectively used in medical diagnostic support tools to increase the diagnostic accuracy and to provide additional knowledge to medical stuff. Effects of principle component analysis on the assessment of exercise stress test with support vector machine in determination of coronary artery disease are studied in this work. Study dataset consist of 480 patients with 23 features for each patient. By reducing study dataset with principle component analysis method, optimum support vector machine model is found for each reduced dimension. According to the obtained results, optimum support vector machine model in which the dataset is reduced to 18 features with principle component analysis is more accurate than optimum support vector machine model which uses the whole 23 featured dataset. Besides, principle component analysis implementation decreases the training error and the sum of the training and test times.

A color image watermarking scheme based on artificial immune recognition system

This study suggests a novel watermarking technique that uses artificial immune recognition system to protect color images intellectual property rights. The watermark is embedded in the blue channel of a color image. m-bit binary sequence embedded into the color image is used to train artificial immune recognition system. With this composed technique, extracting the watermark which is embedded into the color image is carried out using artificial immune recognition system. It is observed that the composed technique achieves high performance to process of extracting this watermark. The watermark is extracted successfully from the watermarked image after various image processing attacks as well.

Noise robust sound event classification with convolutional neural network

Abstract Automatic sound recognition (ASR) is a remarkable field of research in recent years. The ability to automatically recognize sound events through computers in a complex audio environment is very useful for machine hearing, acoustic surveillance and multimedia retrieval applications. On the other hand, ASR task become highly difficult as the ambient noise levels increase and many traditional methods show very weak performance under noise. Recent studies has shown that spectrogram image features (SIF) have high performance under noise, while success rates in clean conditions are relatively lower than in the state-of-the-art approaches. In this study, after converting highly overlapped spectrograms into linear quantized images and reducing dimensions by applying various image resizing methods, feature extraction and classification are performed with convolutional neural networks (CNN), which have very high performance in image classification. In the mismatched case, the proposed method achieves a performance improvement of 4.5%, which is equivalent to a relative error reduction of 63.4%, with a classification success of 97.4%, while the multicondition training method achieves an average of 98.63% success rate.

Watermarking schema using an artificial immune system in spatial domain

In this paper, we propose a robust watermarking method for image copyright protection in spatial domain based on artificial immune system (AIS). Our method optimizes robustness and imperceptibility which are known to be inversely proportional to each other. The robustness of our watermark method was tested extensively against attacks by lossy JPEG compression. Performance of proposed method was evaluated by comparing with the genetic watermarking method and also the randomly chosen block based watermark method. The experiments indicate that our method is better at robustness over genetic watermarking method. We have shown that our method improves the robustness and imperceptibility comparing with the randomly chosen block based watermarking.

Effects of discretization on determination of coronary artery disease using support vector machine

In this paper, the effect of discretization on determination of coronary artery disease using exercise stress test data by support vector machine classification method is investigated. The study dataset is obtained from cardiology department of Meram faculty of medicine including 480 patients having 23 features. Four classification models are composed. In the first model, the data is classified simply by normalizing it into [-1,1] range. In the second, third and fourth models, the data is classified by employing entropy-MDL, equal width and equal frequency discretization methods on it respectively. Support vector machine is used as the classifier for all classification models. The results show that classification performance of the model implemented by entropy-MDL discretization has the best value.

Color Image Watermarking Scheme Based on Efficient Preprocessing and Support Vector Machines

This paper suggests a new block based watermarking technique utilizing preprocessing and support vector machine (PPSVMW) to protect color image’s intellectual property rights. Binary test set is employed here to train support vector machine (SVM). Before adding binary data into the original image, blocks have been separated into two parts to train SVM for better accuracy. Watermark’s 1 valued bits were randomly added into the first block part and 0 into the second block part. Watermark is embedded by modifying the blue channel pixel value in the middle of each block so that watermarked image could be composed. SVM was trained with set-bits and three other features which are averages of the differences of pixels in three distinct shapes extracted from each block, and hence without the need of original image, it could be extracted. The results of PPSVMW technique proposed in this study were compared with those of the Tsai’s technique. Our technique was proved to be more efficient.

Human Gesture Recognition using Keyframes on Local Joint Motion Trajectories

Human Action Recognition (HAR) systems are systems that recognize and classify the actions that users perform against the sensor or camera. In most HAR systems, an input test data is compared with the reference data in the database using various methods. Classification process is performed according to the result obtained. The size of the test or reference data directly affects the operation speed of the system. Reduced data size allows a significant performance increase in system operation speed. In this study, action recognition method is proposed by using skeletal joint information obtained by Microsoft Kinect sensor. Splitting keyframes are obtained from the skeletal joint information. The keyframes are observed as a distinguishing feature. Therefore, these keyframes are used for the classification process. Keeping the keyframes instead of keeping the position or angle information of action in the reference database can benefit from memory and working time. The weight value of each keyframes is calculated in the method. The problem of temporal differences that occur when comparing test and reference action is solved by Dynamic Time Warping (DTW). The k-nearest neighbor’s algorithm is used for classification according to the obtained results from DTW. The sample has been tested in a data set so that the success of the method can be tested. As a result, 100% correct classification was achieved. It is also suitable for working at real time systems. Breakpoints can also be used to provide feedback to the user as a result of the classification process. The magnitude and direction of the keyframes, the change in the trajectory of joint, the position and the time of its existence also give information about the time errors.

Using Chaotic System in Encryption

In this paper chaotic systems and RSA encryption algorithm are combined in order to develop an encryption algorithm which accomplishes the modern standards. E.Lorenz’s weather forecast’ equations which are used to simulate non-linear systems are utilized to create chaotic map. This equation can be used to generate random numbers. In order to achieve up-to-date standards and use online and offline status, a new encryption technique that combines chaotic systems and RSA encryption algorithm has been developed. The combination of RSA algorithm and chaotic systems makes encryption system.