Yazar "İncetaş, Mürsel Ozan" seçeneğine göre listele
Listeleniyor 1 - 7 / 7
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe A novel image Denoising approach using super resolution densely connected convolutional networks(2022) İncetaş, Mürsel Ozan; Uçar, Mehmet; Uçar, Emine; Köse, UtkuImage distortion effects, called noise, may occur due to various reasons such as image acquisition, transfer, and duplication. Image denoising is a preliminary step for many studies in the field of image processing. The vast majority of techniques in the literature require parameters that the user must determine according to the noise intensity. Due to the user requirement, the developed techniques become almost impossible to use by another computer system. Therefore, the Densely Connected Convolutional Networks structure-based model is proposed to remove noise from gray-level images with different noise levels in this study. With the developed approach, the obligation of the user to enter any parameters has been eliminated. For the training of the proposed method, 2200 noisy images with 11 different levels derived from the BSDS300 Train dataset (original 200 images) were used, and the success of the method was evaluated with 1100 noisy images derived from the BSDS300 Test dataset (original 100 images). The images used to evaluate the success of the proposed method were compared to both the traditional and state-of-the-art techniques. It was observed that the average SSIM / PSNR values obtained with the proposed method for the whole test dataset were 0.9236 / 33.94 at low noise level (sigma(2) = 0.001) and 0.7156 / 26.39 at high noise level (sigma(2) = 0.020). The results show that the proposed method is a very effective and efficient noise filter for image denoising.Öğe A Stacking Ensemble Learning Approach for Intrusion Detection System(2021) Uçar, Murat; Uçar, Emine; İncetaş, Mürsel OzanIntrusion detection systems (IDSs) have received great interest in computer science, along with increased network productivity and security threats. The purpose of this study is to determine whether the incoming network traffic is normal or an attack based on 41 features in the NSL-KDD dataset. In this paper, the performance of a stacking technique for network intrusion detection was analysed. Stacking technique is an ensemble approach which is used for combining various classification methods to produce a preferable classifier. Stacking models were trained on the NSLKDD training dataset and evaluated on the NSLKDDTest+ and NSLKDDTest21 test datasets. In the stacking technique, four different algorithms were used as base learners and an algorithm was used as a stacking meta learner. Logistic Regression (LR), Decision Trees (DT), Artificial Neural Networks (ANN), and K Nearest Neighbor (KNN) are the base learner models and Support Vector Machine (SVM) model is the meta learner. The proposed models were evaluated using accuracy rate and other performance metrics of classification. Experimental results showed that stacking significantly improved the performance of intrusion detection systems. The ensemble classifier (DT-LR-ANN + SVM) model achieved the best accuracy results with 90.57% in the NSLKDDTest + dataset and 84.32% in the NSLKDDTest21 dataset.Öğe Anisotropic diffusion filter based on spiking neural network model(2022) İncetaş, Mürsel OzanImage denoising is one of the most important steps in image processing. Anisotropic diffusion filters (ADFs), which are quite popular, stand out with their edge preservation properties, as well as their denoising success. However, ADFs determine edge pixels using the gradient value. In recent years, it has been observed that bioinspired studies based on the human visual system have yielded successful edge detection results. In this study, a new spiking neural network-based approach using the conductance-based integrate and fire neuron model is presented for the calculation of the ADF diffusion coefficient. The success of the proposed method was tested with 1100 noisy images derived from the BSDS300 Test dataset (100 original images) by adding additive white Gaussian noise. The SSIM and PSNR results showed that the proposed method is a very effective and efficient denoising filter.Öğe Automatic color edge detection with similarity transformation(2019) İncetaş, Mürsel Ozan; Demirci, Recep; Yavuzcan, H. GüçlüEdge detection is an important step in image processing. As edge is intensity variation with spatial coordinates, the similarities between neighboring pixels could be used for edge detection. It has been observed that the effective results could be attained by thresholding the homogeneity images generated by means of the similarity transformation. Nevertheless, the user-defined normalization coefficient in similarity transform stage seriously effects edge detection performance and it needs to be automatically selected for every particular image. In this study, a new approach in which the normalization coefficient is automatically determined has been presented. The automating process of the similarity transform has been performed according to the gray level values of the neighboring pixels. The gray level differences of the central pixel and other neighboring pixels have been used to determine the similarity coefficient. Subsequently, the binarization process of the homogeneity images obtained with proposed algorithm have been completed with different thresholding techniques. Additionally, the F-score of the proposed edge detection has been obtained with 200 images in the BSDS training dataset. The achieved F-score values have showed that the performance of automatic approach is quite high.Öğe Edge detection using integrate and fire neuron model(2019) İncetaş, Mürsel Ozan; Arslan, Rukiye UzunEdge detection is one of the most basic stages of image processing and have been used in many areas. Its purpose is to determine the pixels formed the objects. Many researchers have aimed to determine objects' edges correctly, like as they are determined by the human eye. In this study, a new edge detection technique based on spiking neural network is proposed. The proposed model has a different receptor structure than the ones found in literature and also does not use gray level values of the pixels in the receptive field directly. Instead, it takes the gray level differences between the pixel in the center of the receptive field and others as input. The model is tested by using BSDS train dataset. Besides, the obtained results are compared with the results calculated by Canny edge detection method.Öğe Effects of character recognition with shell histogram method using plate characters(2019) Arslan, Rukiye Uzun; Dikici, Sedat; İncetaş, Mürsel OzanCharacter recognition is a study that has been used in various fields for many years. In character recognition, the aim is to identify the various texts, letters and symbols in the images as accurately and quickly as possible. In addition to the Optical Character Recognition (OCT) method, which is used as a very common method, there are many feature extraction methods in which character image features are compared. In this study, which is presented as another feature extraction method, the letters on the license plates are recognized. The characters were examined using the circular shape histogram technique and histograms were obtained from the sectors within the circular regions. Feature vectors for letter characters were created using character pixel densities in sectors. Feature vectors are analyzed linearly and an alternative quick character recognition method is presented. With the proposed method, the element numbers of the feature vectors are kept constant. In this way, both the processing speed is increased and the processing speed variations are minimized. The results show that the proposed method requires lesser parameters than the OCT method, but also has a significant success rate according to known feature extraction methods.Öğe Using Machine Learning Algorithms for Jumping Distance Prediction of Male Long Jumpers(2022) İncetaş, Mürsel Ozan; Uçar, Murat; Bayraktar, Işık; Çilli, MuratThe long jump is defined as an athletic event, and it has also been a standard event in modern Olympic Games. The purpose of the athletes is to make the distance as far as possible from a jumping point. The main purpose of this study was to determine the most successful machine learning algorithm in the prediction of the long jump distance of male athletes. In this paper, we used age and velocity variables for predicting the long jump performance of athletes. During the research, 328 valid jumps belonging to 73 Turkish male athletes were used as data. In determining the most successful algorithm, mean absolute error (MAE), root mean square error (RMSE), Mean Squared Error (MSE), R2 score, Explained Variance Score (EVS), and Mean Squared Logarithmic Error (MSLE) values were taken into consideration. The outcomes of the analysis showed that long jump performance can be determined by chosen independent variables. The 5-fold cross-validation technique was used for the performance evaluation of the models. As a result of the experimental tests, the Gradient Boosting Regression Trees (GBRT) algorithm reached the best result with an MSE value of 0.0865. In this study, it was concluded that the machine learning approach suggested can be used by trainers to determine the long jump performance of male athletes.
