Search

Buçalışmada EMG sinyalleri kullanılarak 14 farklı parmak hareketi önerilen yenibir öznitelik çıkarma yöntemiyle sınıflandırılmıştır. EMG sinyallerinden parmakhareketlerinin tespiti / sınıflandırılması, ön işleme, özellik çıkarma vesınıflandırma adımlarını içeren 3 ana adımdan oluşur. EMG sinyallerininsınıflandırılmasında, sınıflandırıcının performansı doğrudan öznitelik çıkarmayöntemine bağlıdır. Bu sebeple öznitelik çıkarma yöntemi uygun bir biçimdeseçilmelidir. Literatürde zaman, histogram ve frekans tabanlı birçok öznitelikçıkarma yöntemi vardır. Ancak bu yöntemler yüksek zaman karmaşıklığı, çok fazlaişleme ihtiyaç duyma, fazla sayıda kontrol parametresi bulundurma, vb. birçokdezavantaja sahiptir. Bu çalışmada EMG sinyallerinden çeşitli parmakhareketlerinin sınıflandırılması amacıyla sayılan dezavantajları ortadankaldırmak için yeni bir öznitelik çıkarma yöntemi sunulmuştur. Bu yöntem ikizaman tabanlı öznitelik çıkarma yönteminin hibritleştirilmesiyle eldeedilmiştir. 10 kat çapraz doğrulamayla elde edilen deneysel sonuçlarına göre, önerilenyöntemin %97,48 doğruluk oranıyla bu çalışmada kullanılan diğer 9 öznitelikçıkarma yöntemlerinden daha iyi olduğu deneysel olarak gösterilmiştir. Deneyselsonuçlar istatistiksel yöntemlerle desteklenmiştir., This studyaims to discuss classification of 14 different finger movements from EMG signalsby using new feature extraction technique. The detection/classification of finger movements consists of 3 mainsteps including, preprocessing, feature extraction and classification steps. Inclassification of EMG signals, the performance of the classifier directly dependson feature extraction methods, including, time, histogram and frequency-basedmethods. However, these feature extraction methods have several drawbacksincluding, high time complexity, high computation demand, user suppliedparameters, etc. In this paper, a new feature extraction method has beenproposed for the classification of finger movements from EMG signals toovercome these problems. The proposed method based on hybridization of 2-timedomain feature extraction techniques. The use of this method resulted in an accuracyof 97.48% after 10-fold-cross-validation. The experimental results supportedwith statistical analysis show that proposed method is better than 9 featureextraction methods investigated in this paper

Date of Conference: 16-19 May 2016 Conference Name: IEEE 24th Signal Processing and Communications Applications Conference, SIU 2016 In this paper, we study multi-class classification of tweets, where we introduce highly efficient dimensionality reduction techniques suitable for online processing of high dimensional feature vectors generated from freely-worded text. As for the real life case study, we work on tweets in the Turkish language, however, our methods are generic and can be used for other languages as clearly explained in the paper. Since we work on a real life application and the tweets are freely worded, we introduce text correction, normalization and root finding algorithms. Although text processing and classification are highly important due to many applications such as emotion recognition, advertisement selection, etc., online classification and regression algorithms over text are limited due to need for high dimensional vectors to represent natural text inputs. We overcome such limitations by showing that randomized projections and piecewise linear models can be efficiently leveraged to significantly reduce the computational cost for feature vector extraction from the tweets. Hence, we can perform multi-class tweet classification and regression in real time. We demonstrate our results over tweets collected from a real life case study where the tweets are freely-worded, e.g., with emoticons, shortened words, special characters, etc., and are unstructured. We implement several well-known machine learning algorithms as well as novel regression methods and demonstrate that we can significantly reduce the computational complexity with insignificant change in the classification and regression performance.

In this paper we propose a novel framework using features extracted from Wikipedia for the task of matching named entities. We present how the employed features are extracted from Wikipedia and how its structure is utilized. We describe how a term-concepts table constructed from Wikipedia and the redirect links are integrated in the framework. In addition, the internal links within Wikipedia along with the categories structure are also used to compute the relatedness between concepts. We evaluate the built framework and report its performance in the applications of word sense disambiguation and named entities matching. The system performance is compared against other learning-based state-of-the-art systems and its reported results are found to be competitive. We also present a method in this paper for named entities matching that is context-independent and compare its results with those of other systems.

6th International Conference on Computer Science and Engineering, UBMK 2021 -- 15 September 2021 through 17 September 2021 -- -- 176826, Nowadays, shopping is done more comfortably and without time constraints with the throwing of e-commerce platforms. These platforms allow consumers to examine reviews before purchasing products. Thus, consumers can decide whether to buy a product with positive or negative comments about the products. In this paper, Turkish sentiment analysis was carried out on the product comments at the Hepsiburada platform. For sentiment analysis, firstly, the success of Random Forest, Naive Bayes and Logistic Regression machine learning methods was measured. Then, the effect of BERT, ELECTRA and ALBERT language models on sentiment analysis was analyzed and the success of language models was compared with machine learning methods. While Naive Bayes achieved the highest accuracy with 89.95% among machine learning methods, ELECTRA was the most successful with 92.54% among language models. As a result of the study, it has been shown that the ELECTRA and ALBERT language models are more successful than machine learning methods. © 2021 IEEE

Sentiment analysis is a research field that aims to identify the sentiment orientation (as, positive, negative or neutral) of a particular topic in text documents through machine learning, statistics and natural language processing techniques. Convolutional neural networks are a type of deep learning methods, which process data with a grid-like topology. In this paper, we present empirical results for three deep learning architectures based on convolutional neural network for sentiment analysis on Turkish. The first architecture initially employs word-embedding schemes to represent text documents. Then, a stack of convolution layers (i.e., 1-gram, 2-gram and 3-gram) has been employed to extract 1-gram, 2-gram and 3-gram based features. For each layer, constant number of filters have been employed to construct feature maps. The second examined architecture employs recurrent convolution and maximum pooling schemes on word embedding based representation. The third architecture is a convolution based pyramid configuration. To represent text corpus, word2vec, fastText, GloVe and LDA2vec word embedding schemes have been utilized. The empirical results on Turkish sentiment classification indicate that convolutional deep learning based architectures outperform conventional machine learning methods (such as, k-nearest neighbor algorithm, support vector machines, logistic regression and Naïve Bayes algorithm) and conventional deep learning architectures (such as, recurrent neural networks, long short term memory architecture and gated recurrent unit). We obtained a classification accuracy of 92.53% with convolutional neural network based architecture in conjuncton with word2vec (skip-gram model) based word embedding scheme., Duygu analizi (görüş madenciliği), metin belgeleri içerisinde yer alan nesnelere, ürünlere, servislere ya da organizasyonlara ilişkin görüş, duygu, tutum gibi öznel bilgilerin, makine öğrenmesi, istatistik ve doğal dil işleme gibi alanlardan teknik ve yöntemlerin kullanılması ile çıkarılmasını amaçlayan bir araştırma alanıdır. Duygu analizi, yapısal olmayan bilgiden, yapısal, anlamlı ve kullanışlı bilgiler çıkarılmasını olanaklı hale getirir. Bu bilgi, karar destek sistemleri ve bireysel karar vericiler için önemli bir kaynak olarak işlev görür. Evrişimli sinir ağları, veriyi ızgara benzeri bir topoloji ile işleyen bir tür derin öğrenme yöntemidir. Bu çalışmada, Türkçe metin belgeleri üzerinde duygu analizi için, evrişimli sinir ağı tabanlı üç temel derin öğrenme mimarisinin etkinliği değerlendirilmektedir. Çalışma kapsamında önerilen birinci mimaride, gömme katmanında, metin belgesinde yer alan kelimeler için, kelime gömme yöntemleri tabanlı temsil elde edilmektedir. Ardından, evrişim katmanları yığını (1-gram, 2-gram ve 3-gram) kullanılarak 1-gram, 2-gram ve 3-gram tabanlı özniteliklerin çıkarımı gerçekleştirilmektedir. Her bir katmanda, öznitelik haritalarının oluşturulması için sabit sayıda 80 filtre uygulanmaktadır. İncelenen ikinci evrişimli sinir ağı tabanlı mimaride gömme katmanı sonucu elde edilen metin temsili, yinelenen evrişim ve maksimum havuzlama katmanlarına tabi tutulmaktadır. İncelenen üçüncü mimari ise, evrişim tabanlı piramit mimarisidir. Metin belgelerinin temsilinde, word2vec, fastText, GloVe ve LDA2vec olmak üzere dört temel kelime gömme yöntemi incelenmektedir. İncelenen evrişimli sinir ağı tabanlı mimarilerin, Türkçe duygu analizi için, geleneksel makine öğrenmesi sınıflandırıcılarına (k-en yakın komşu algoritması, destek vektör makineleri, lojistik regresyon ve Naive Bayes algoritması) ve temel derin öğrenme mimarilerine (tekrarlayan sinir ağları, uzun kısa süreli bellek birimleri ve geçitli tekrarlayan birim) kıyasla daha yüksek başarım elde ettiği görülmektedir. Word2vec (Skip-gram modeli) kelime kodlaması yöntemi ile evrişimli sinir ağı tabanlı mimari ile %92.53 doğru sınıflandırma oranı elde edilmiştir.

In this paper, syntactic derivation of the correlative constructions in modern Turkish is examined within the framework of the Generative Grammar for the first time. The following points are emphasized in the study: (i) for simple correlative constructions, a subordinating correlative clause containing a wh-phrase must be merged to the demonstrative in the matrix clause as local as possible in a structural manner, (ii) there are two plausible syntactic accounts for generating Turkish simple correlative constructions, i.e. the first is based on base-generated approach and the second is based on movement approach, (iii) the derivation of simple correlative constructions is diffrent from the one in unconditionals, (iv) regarding multiple correlative constructions, the most local position for a subordinate clausal material which should be merged to the matrix is the minimal IP containing demonstrative pronouns, and (v) the multiple correlative constructions in Turkish can be generated using the non-movement IP adjunction strategy., Bu çalışmada, Türkçedeki bağlılaşık yapıların sözdizimsel türetimi Üretici Dilbilgisi Kuramı çerçevesinde ilk kez incelenecektir. Çalışmada genel olarak şu hususlar vurgulanacaktır: (i) basit bağlılaşık yapılarda ne-öbeğini içeren bağlılaşık yan tümce, ilişkilendirildiği bağlılaşık ana tümce içindeki işaret sözcüğüne yapısal olarak mümkün olan en yerel düzende bağlanmalıdır, (ii) basit bağlılaşık yapıların üretimi için biri taban-üretimli diğeri taşıma içeren iki farklı yaklaşım mümkündür, (iii) basit bağlılaşık yapıların üretilme biçimi şartsız yapılardan farklıdır, (iv) çoklu bağlılaşık yapılarda bağlılaşık yan tümcenin ana tümceye bağlanabileceği en yerel konum işaret sözcüklerini barındıran en küçük çekim öbeğine eklentilemedir ve (v) Türkçedeki çoklu bağlılaşık tümceler taşıma içermeyen çekim öbeği eklentilemesi yapısıyla üretilebilmektedir.

Otomatik görüntü altyazısı, yapay zekânın hem bilgisayarlagörme hem de doğal dil işleme alanlarını kapsayan bir konudur. Makine çevirisialanındaki gelişmelerden ilham alan ve bu alanda başarılı sonuçlar verenkodlayıcı-kod çözücü tekniği, şu anda İngilizce görüntü altyazısı konusunda kullanılanmevcut yöntemlerden biridir. Bu çalışmada, Türkçe dili için otomatik görüntü altyazısıoluşturan bir model sunulmaktadır. Bu çalışma, verilen görüntülerinözelliklerini çıkarmaktan sorumlu olan bir CNN kodlayıcıyı, altyazı oluşturmaktansorumlu olan bir RNN kod çözücüsü ile birleştirilerek, Türkçe MS COCO veritabanını üzerinde Türkçe görüntü altyazısı kodlayıcı-kod çözücü modelini test etmektedir.Üretken modelin performansı yeni oluşturulan veri tabanında hem BLEU, METEOR,ROUGE ve CIDEr gibi en yaygın değerlendirme ölçütleri hem de insan tabanlıyöntemler kullanılarak değerlendirilmiştir. Sonuçlar, önerilen modelinperformansının hem niteliksel hem de niceliksel olarak tatmin edici olduğunugöstermektedir. Çalışma sonunda hazırlanan, herkesin kullanımına açık bir Web uygulamasıuygulaması[1]sayesinde Türkçe dili için MS COCO görüntülerine ait Türkçe girişlerinyapıldığı bir ortam kullanıcıya sunulmuştur. Tüm görüntüler tamamlandığında,Türkçe’ye özgü ve karşılaştırmalı çalışmaların yapıldığı bir veri kümesitamamlanmış olacaktır.[1] http://mscococontributor.herokuapp.com/website/, Automatic image captioning is a challenging issue inartificial intelligence, which covers both the fields of computer vision andnatural language processing. Inspired by the later advances in machinetranslation, a successful encoder-decoder technique is currently thestate-of-the-art in English language captioning. In this study, we proposed animage captioning model for Turkish Language. This paper evaluate theencoder-decoder model on MS COCO database by coupling an encoder CNN -thecomponent that is responsible for extracting the features of the given images-,with a decoder RNN -the component that is responsible for generating captionsusing the given inputs- to generate Turkish captions. We conducted theexperiments using the most common evaluation metrics such as BLEU, METEOR,ROUGE and CIDEr. Results show that the performance of the proposed model issatisfactory in both qualitative and quantitatively evaluations. Finally, thisstudy introduces a Web platform, which is proposed to improve the dataset viacrowd-sourcing and free to use. The Turkish MS COCO database is available forresearch purpose. When all the imagesare completed, a Turkish dataset will be available for comparative studies.

© 2018 IEEE.Image segmentation is an important image processing method. The segmentation method, which is a process of dividing into different regions, is a meaningful according to certain features of the image. In this paper, histogram based, edge based, region based and active contour methods most widely used in image segmentation methods are investigated. Advantages and disadvantages of these methods are explained and the effects of these methods on the image are compared.

Doğrudan pazarlama, uygun ürünleri uygun kişilerle en kısa yoldan buluşturma sürecidir. Son yılların en popüler pazarlama yaklaşımlarından birisidir. Bu çalışmada turizm sektörüne ait isimsizleştirilmiş bir veri tabanını kullandık. Bir otel zinciri için yapılan kampanya kapsamında veri madenciliği tekniklerini uygulayarak hedef kitle seçimi yaptık. Çalışmada birçok makine öğrenmesi yöntemini denedik. Sonuç olarak; geçmişte yapılan ve herhangi bir makine öğrenmesi yöntemi kullanılmadan hazırlanan kampanya sonuçlarına göre daha iyi sonuçlar elde ederken benzer analizlerde kullanılabilecek bir altyapı oluşturmuş olduk. Direct marketing, which is one of the most popular marketing approaches in the recent years, can be defined as the shortest and most accurate way of matching suitable products with the appropriate people. In this paper, we used an anonymous tourism sector database. We applied data mining techniques for target audience selection for a hotel campaign. As a result; we achieved better results than the results of previous campaigns without the use of any machine learning methods and we created an infrastructure that can be used for similar analyses. Publisher's Version

Date of Conference: 2-5 May 2018 In this paper, we study online nonlinear data regression and propose a highly efficient long short term memory (LSTM) network based architecture. Here, we also introduce on-line training algorithms to learn the parameters of the introduced architecture. We first propose an LSTM based architecture for data regression. To diminish the complexity of this architecture, we use an energy efficient operator (ef-operator) instead of the multiplication operation. We then factorize the matrices of the LSTM network to reduce the total number of parameters to be learned. In order to train the parameters of this structure, we introduce online learning methods based on the exponentiated gradient (EG) and stochastic gradient descent (SGD) algorithms. Experimental results demonstrate considerable performance and efficiency improvements provided by the introduced architecture.

Date of Conference: 2-5 May 2018 In this paper, we investigate online parameter learning for Long Short Term Memory (LSTM) architectures in distributed networks. Here, we first introduce an LSTM based structure for regression. Then, we provide the equations of this structure in a state space form for each node in our network. Using this form, we then learn the parameters via our Distributed Particle Filtering based (DPF) training method. Our training method asymptotically converges to the optimal parameter set provided that we satisfy certain trivial requirements. While achieving this performance, our training method only causes a computational load that is similar to the efficient first order gradient based training methods. Through real life experiments, we show substantial performance gains compared to the conventional methods.

Mobile devices begin to spread increasingly recently to offer a lot of services which personal computers offer. This condition has led to increase in the number of security threats in mobile devices and services. In this paper, it has been made a research on mobile malware and malware detection techniques. Within the scope of the study, a a permission based detection system based on the machine learning methods for Android malware was developed. Developed system is analyzed by using Random Forest, Support Vector Machine and Artificial Neural Networks algorithms.

In this paper, an object detection approach suitable for mobile augmented reality is presented. The baseline approach is bused on matching keypoint descriptors and yerin.,ing these matches with geometric constraints. The performance optimizations necessary for speeding up matching are detailed. It is [ifs demonstrated that it is possible to increase the performance of the Locality Sensitive Hashing by exploiting approaches from the information retrieval field.

Nowadays, along with the development of information technologies, storage and analysis of biomedical datasets are easy in health sector. In this area, Machine Learning methods provide a great contribution for evaluation and interpretation of data. In this paper, in addition to Support Vector Machines, Decision Tree, K-Nearest Neighbors, Naive Bayes and Dictionary Learning methods, Random Feature Subspaces (RDL) and Random Instance Subspaces (BDL) methods which are the ensembles of Dictionary Learning are used in biomedical data classification. In the test results, SVM and Dictionary Learning methods, RDL and BDL, which are generated using random feature/instance subspaces achieve optimum accuracy results.

25th Signal Processing and Communications Applications Conference (SIU) -- MAY 15-18, 2017 -- Antalya, TURKEY WOS: 000413813100575 Large scale dictionaries such as wordnets are one of the important structures used in natural language processing. Wordnet is a comprehensive dictionary containing semantic relations including synonyms, antonyms, etc. In this paper, we try to extract Hypernym-Hyponym relations which are one of the main parts of WordNet. For this aim, we used Wikipedia, Contemporary Dictionary of Turkish and Wiktionary as corpora. We used Finite State Machines for developing several rules for Hypernym-Hyponym extraction. Turk Telekom, Arcelik A S, Aselsan, ARGENIT, HAVELSAN, NETAS, Adresgezgini, IEEE Turkey Sect, AVCR Informat Technologies, Cisco, i2i Syst, Integrated Syst & Syst Design, ENOVAS, FiGES Engn, MS Spektral, Istanbul Teknik Univ

25th Signal Processing and Communications Applications Conference (SIU) -- MAY 15-18, 2017 -- Antalya, TURKEY WOS: 000413813100251 The Internet and social media provide a major source of information about people's opinions. Due to the rapidly growing number of online documents, it becomes both time-consuming and hard task to obtain and analyze the desired opinionated information. Sentiment analysis is the classification of sentiments expressed in documents. To improve classification perfromance feature selection methods which help to identify the most valuable features are generally applied. In this paper, we compare the performance of four feature selection methods namely Chi-square, Information Gain, Query Expansion Ranking, and Ant Colony Optimization using Maximum Entropi Modeling classification algorithm over Turkish Twitter dataset. Therefore, the effects of feature selection methods over the performance of sentiment analysis of Turkish Twitter data are evaluated. Experimental results show that Query Expansion Ranking and Ant Colony Optimization methods outperform other traditional feature selection methods for sentiment analysis. Turk Telekom, Arcelik A S, Aselsan, ARGENIT, HAVELSAN, NETAS, Adresgezgini, IEEE Turkey Sect, AVCR Informat Technologies, Cisco, i2i Syst, Integrated Syst & Syst Design, ENOVAS, FiGES Engn, MS Spektral, Istanbul Teknik Univ

Semantic place prediction problem is the process of giving semantic names, such as school, home, office, to locations. Different than the localization problem where the coordinates of a place are predicted, the aim is to semantically characterize the location. While the GPS coordinates of a place can be utilized in solving the problem, phone usage patterns of the users in that location can be used as well. In this paper, the aim is to semantically classify locations of smart phone users utilizing the data collected from wireless interfaces and the motion sensors available on the phones with machine learning algorithms. The efficiency of features extracted from raw data is analysed in terms of metrics such as accuracy, using different classification algorithms. The results show that, while random forest and decision tree algorithms achieve 66% accuracy with only temporal features, adding features from device properties and activity features increases the accuracy up to 99%.

Artificial Bee Colony algorithm, inspired by the foraging behavior of real honey bees, is one of the most important swarm intelligence based optimization algorithms. Like other population based evolutionary computation techniques, Artificial Bee Colony algorithm is suitable for parallelization on distributed architectures. In this paper, we presented a new emigrant creation strategy that is being distributed between subcolonies running simultaneously on the independent compute nodes. The running times and objective function values obtained by the parallelized Artificial Bee Colony algorithm with the proposed model on different number of compute nodes are compared with the sequential counterpart of the algorithm and it is seen that convergence performance of the parallelized Artificial Bee Colony algorithm is significantly improved with the proposed emigrant creation strategy.

In this paper, Discrete-wavelet transform (DWT) based algorithm are used for speech signal denoising. Both hard and soft thresholding are used for denoising. Analysis are done on noisy speech signal corrupted by AGWN(Additive gaussian white noise) noise. Simulation and results are performed by using SignalLab application which developed for this research. MSE and SNR is used to compare results. Qt/C++ IDE is used to develop the application. This application developed for not only analyzing speech signal, but also can be used in an embedded system or a real-time applications. According to the obtained results, DWT is a good method for speech denoising when parameters are selected correctly.

24th Signal Processing and Communication Application Conference (SIU) -- MAY 16-19, 2016 -- Zonguldak, TURKEY, WOS: 000391250900110, In this paper, 11 different power quality disturbances were automatically detected by using statistical features with wavelet transform and norm entropy techniques. The best of the created features were selected with forward selection algorithm. Performance of classification algorithms, Support Vector Machines (SVM), Multi Layer Perceptron (MLP), k Nearest Neighbor (KNN) and random subspace KNN (Sub-KNN) which is an ensemble method, were examined. Consequently, the best classification accuracy of 99.3% was achieved by using Sub-KNN and it was appeared that compared to other methods, this algorithm was more robust against the noise., IEEE, Bulent Ecevit Univ, Dept Elect & Elect Engn, Bulent Ecevit Univ, Dept Biomed Engn, Bulent Ecevit Univ, Dept Comp Engn

In this paper, an effective classification approach to classify harmonic data has been proposed. In the proposed classifier approach, harmonic data obtained through a 3-phase system have been classified by using $k$-means and least square support vector machine (LS-SVM) models. In order to obtain class details regarding harmonic data, a $k$-means clustering algorithm has been applied to these data first. The training of the LS-SVM model has been realized with the class details obtained through the $k$-means algorithm. To increase the efficiency of the LS-SVM model, the regularization and kernel parameters of this model have been determined with a grid search method and the training phase has been realized. Backpropagation neural network and J48 decision tree classifiers have been applied to the same data and results have been obtained for the purpose of comparing the performance of the LS-SVM model. The real data obtained from the output of distribution system have been used to assess the performance of the proposed classifier system. The obtained results and comparisons suggest that the proposed classifier system approach is quite efficient at classifying harmonic data.

The probabilistic neural network (PNN) is a special type of radial basis neural network used mainly for classification problems. Due to the size of the network after training, this type of network is usually used for problems with a small-sized training dataset. In this paper, a new training algorithm is presented for use with large training databases. Application to the handwritten digit database shows that the reduced PNN performs better than the standard PNN for all of the studied cases with a big gain in size and processing speed. This new type of neural network can be used easily for problems with large training databases like biometrics and data mining applications. An extension of the network is possible for new training samples and/or classes without retraining.

23nd Signal Processing and Communications Applications Conference (SIU) -- MAY 16-19, 2015 -- Inonu Univ, Malatya, TURKEY WOS: 000380500900192 Superpixels recently gained in importance in image segmentation and classification problems. In scene labeling the image is initially segmented into visually consistent small regions using a superpixel algorithm; then, superpixels are parsed into different classes. Classification performance heavily depends on the properties and parametric settings of the superpixel algorithm in use. In this paper, a method is proposed to improve scene labeling accuracy by fusing at classifier level the results of multiple superpixel segmentations. First, likelihood ratios are determined for superpixel labels using simple, nonparametric SuperParsing algorithm, which requires no training. Then, final scene segmentation and labeling is performed by pixel-level fusion of the likelihood ratios that are computed for alternative superpixel segmentation scenarios. The proposed method is tested on the SIFT Flow dataset consisting of 2,688 images and 33 labels, and is shown to outperform SuperParsing in terms of classification accuracy. Dept Comp Engn & Elect & Elect Engn, Elect & Elect Engn, Bilkent Univ

Image denoising is an important task in image processing. Various filtering techniques has been developed for image denoising but the algorithms have specific limitations. The main feature of a good image denoising technique is that it will remove noise while preserving structure of image. In this paper, we present a new image denoising method based on averaging. The denoising performance is measured according to PSNR (Peak Signal-to-Nosie Ratio) metric by employing experiments. Experimental results show that the proposed method removes noise significantly and preserve image edges.

Speech processing is working area where the speech signal is digitized and processed. In this paper, it is used LSVT (Lee Silverman Voice Treatment) dataset that belonging to the people living speech disorder because of the Parkinson disease. The dataset contains a large number of features. A large number of features can be occurred negative effect on classification because of noise or less important features. Especially, it is suggested that k nearest neighbor is weighted to reduce this effect distance based classifiers. Therefore, it is benefited from Laplacian score to weight features. Consequently, classification accuracy is increased from 73.61% to 85.83% for k nearest neighbor classifier.

One of the well-known recurrent neural networks is the Elman network. Recently, it has been used in applications of system identification. The network has feedforward and feedback connections. It can be trained essentially as a feedforward network by means of the basic backpropagation algorithm, but its feedback connections have to be kept constant. For training success, it is important to select the correct values for the feedback connections. However, finding these values manually can be a lengthy trial-and-error process. This paper investigates the use of the simulated annealing (SA) algorithm to obtain the weight values of both the feedforward and feedback connections of Elman networks used for dynamic system identification. The SA algorithm is an efficient random search procedure, which can simultaneously obtain the optimal weight values of both the feedforward and feedback connections.

In present times, automatic computer interpretation of medical images is of great importance for medical diagnosis. Having previously been an obstacle for medical science, early stage diagnoses together with accurately calculating the size and level of disease are very important elements of today's treatment process. The treatment and diagnosis of brain tumors depend on successful processing and evaluation of magnetic resonance (MR) brain images. Image segmentation is the most intractable process, because of the complicated nature of the MR brain images. In processing and analyzing MR images, there is unsuccessful automatic systems or successful semiautomatic system, which is still required large scale experting involvement and assessment. Within this paper the aim is to propose a system that will automatically extract the location, volume and borders of a brain tumor from patient MRI brain images. To achieve this goal, an automated algorithm has been developed, based upon and derived from a successful semi-automatic hybrid algorithm, for automation of the tumor diagnosis process. Patients' MR brain images from varying stages of the diagnosis and treatment were used for testing and validating the automatic algorithm under development. Based upon the achieved results of this algorithm and method, an automatic, sensitive MR brain computer interpretation can be proposed.

This paper examines the performance of Hidden Markov Tree model based weights in reconstruction quality for an existing task-aware compressive video coding system which aims object detection specifically. The existing system utilizes weights in reconstruction which are computed by tracking of the foreground object. The proposed system acquires similar average PSNR with the existing one which reported some improvement compared to the conventional unweighted reconstruction at low sampling rates. Furthermore, it is a little bit better than the existing system at higher sampling rates. It can be inferred from this study that Bayesian approaches that take account structural dependencies between transformation coefficients has the potential of improving reconstruction quality for such a compressive video coding system with object detection task.

Segmentation of abdominal organs from magnetic resonance data sets is an essential task for several medical procedures and analysis. When this process is done manually, it requires an expert radiologist's time and experience. Classification of organs by using several features for segmentation is a known technique. In recent years, for different application fields it is shown that hierarchical classification systems are more successful than composite feature-single classifier systems. The success of hierarchical systems in medical image segmentation is directly related to the reduction of false positive error in every level of the hierachy. In this paper, 3 different algorithms of Adaboost (Adaptive Boosting) method: Real, gentle and modest are compared in terms of their false positive error performance. Performance metrics are obtained for each organ and the dependence of false positive error on the selected algorithm is examined.

Date of Conference: 20-22 April 2011 Conference Name: 19th Signal Processing and Communications Applications Conference, SIU 2011 In this paper, a new histogram based auto-focusing method for thermal cameras is proposed. This proposed method is realized by FPGA (Field Programmable Gate Array) and DSP (Digital Signal Processor) working together and simultaneously. HF (High Frequency) component, obtained from real-time image flow by FPGA and DSP is used for auto-focusing process. Proposed method is able to determine the focus direction from the HF component produced in the process of histogram equalization by FPGA, unlike Fourier transform and pixel differenve based methods in the literature. With this superiority, proposed method requires no extra calculation for thermal cameras for which histogram equalization is necessary. Analysis show that proposed method is successful on the simulations and scanning thermal cameras.

A powerful method in analysis of large data sets where there are many natural clusters with varying statistics such as different sizes, shapes, density distribution, is the use of sel-forganizing maps (SOMs) [1]. However, further processing tools, such as visualization, interactive clustering, are often necessary to capture the clusters from the learned SOM knowledge. A recent visualization scheme, CONNvis [2], and interactive clustering from CONNvis, utilizes the data topology for SOM knowledge representation by using a weighted Delaunay graph, CONN. In this paper, an automated clustering scheme for SOMs, SOMcluster, which is a two-level clustering of CONN by the skills obtained in the interactive process, is proposed. It is shown that SOMcluster, which does not require the number of clusters a priori, is used successfully for automated segmentation of a remote sensing spectral image which has many clusters some of which were unidentified in previous works.

In stereoscopic 3D video, it is well-known that humans can perceive high quality 3D video provided that one of the views is in high quality. Hence, in stereo video encoding, the best overall rate vs. perceived-distortion performance may be achieved by reduction of the spatial, temporal and/or quantization resolution of the second view, while keeping the first view in full resolution. Objective stereo video quality metric is a crucial point for the success of unequal rate allocation. Since the perceived 3D video quality does not correlate well with the average PSNR of the two views, we proposed a new quantitative measure using a weighted combination of two PSNR values and a jerkiness measure [1]. In this paper, we investigate whether the PSNR weights should be related with spatial activity in different scaling options and video contents. Spatial activity is examined in terms of blockiness and blur. The optimal weights for some specific experiments are determined by matching, i.e. searching the best correlation between, subjective quality test results and the objective measure scores on a set of test videos.

Date of Conference: 9-11 April 2009 Conference Name: IEEE 17th Signal Processing and Communications Applications Conference, SIU 2009 In the last few years, there is great increase in capture and representation of real 3-Dimensonal scenes using 3D animation models. The 3D signals are then compressed, transmitted to the client side and reconstructed for the user view. Each step mentioned here opened a new subject in the field of signal processing. While processing these models, using the model as a whole is not the best approach. Therefore clustering the model vertices became a very common method. For example, it is very common to use motion based clustering in animation compression. In this paper a new dynamic model clustering algorithm is proposed. Animation vertices are first put through PCA and partitioned into their eigenvalues and eigenvectors. The eigenvectors found using the proposed method can be called eigentrajectories. Then the dot product of the these eigentrajectories with the trajectories of the animation vertice are found. These coefficients are used to cluster the animation model. The results and the comparisons with a similar approach show that the proposed algorithm is successful.

Identifying the liver region, calculation of the liver volume and determination of the vessel structure from abdominal computed tomography datasets are some of the essential steps in visualization prior to the hepatic surgery. Because of the high number of slices, manual segmentation of the liver is time consuming, tedious and depends on the experience. On the other hand, the automatic segmentation of the liver is very difficult task because of the gray level similarities of adjacent organs, injection of contrast media and partial volume effect problems. In this paper, we propose an algorithm that can handle these problems. Developed algorithm involves preprocessing, classification and post-processing stages. The proposed method is applied to 17 donor datasets and its performance is evaluated by area error rate calculations.

This paper investigates the control of nonlinear systems by neural networks and fuzzy logic. As the control methods, Gaussian neuro-fuzzy variable structure (GNFVS), feedback error learning architecture (FELA) and direct inverse modeling architecture (DIMA) are studied, and their performances are comparatively evaluated on a two degrees of freedom direct drive robotic manipulator with respect to trajectory tracking performance, computational complexity, design complexity, RMS errors, necessary training time in learning phase and payload variations.