Yazar "Karasulu, Bahadir" seçeneğine göre listele

Listeleniyor 1 - 20 / 22
  • [ X ]
    Öğe
    A case study: People detection and tracking in videos
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter provides three sections. The first section introduces the video database that it is used in our experiments. This database is based on people surveillance footage and called CAVIAR. In our study, the performance results are based on both qualitative and quantitative evaluation. In second section, some experimental results of our study are presented via qualitative and overall quantitative (i.e., numerical) results of performance. In addition, we explain that the capabilities of ViCamPEv are used to obtain these results. In third section, we present both statistical and algorithmic analysis of relevant object D&T methods. These methods are comparable via the results in related tables for related methods. The discussion about given methods are presented through these experimental results and performance evaluation. © The Author(s) 2013.
  • [ X ]
    Öğe
    A hybrid approach based on deep learning for gender recognition using human ear images
    (Gazi Univ, Fac Engineering Architecture, 2022) Karasulu, Bahadir; Yucalar, Fatih; Borandag, Emin
    Nowadays, the use of the human ear images gains importance for the sustainability of biometric authorization and surveillance systems. Contemporary studies show that such processes can be done semi-automatically or fully automatically, instead of being done manually. Due to the fact that deep learning uses abstract features (i.e., representation learning), it reaches quite high performance values compared to classical methods. In our study, a synergistic gender recognition approach based on hybrid deep learning was created based on the use of human ear images in classifying people fully automatically according to their gender. By means of hybridization, hybrid deep neural network architectural models are used, which include both convolutional neural network component and recurrent neural network type components together. In these models, long-short term memory and gated recurrent unit are taken as recurrent neural network type components. Thanks to these components, the hybrid model extracts the relational dependencies between the pixel regions in the image very well. On account of this synergistic approach, the gender classification accuracy of hybrid models is higher than the standalone convolutional neural network model in our study. Two different image datasets with gender marking were used in our experiments. The reliability of the experimental results has been proven by objective metrics. In the conducted experiments, the highest values in gender recognition with hybrid models were obtained with the test accuracy of 85.16% for the EarVN dataset and 87.61% for the WPUT dataset, respectively. Discussion and conclusions are included in the last section of our study.
  • [ X ]
    Öğe
    A Query Evaluation Approach using Opinions of Turkish Financial Market Professionals
    (Assoc Information Communication Technology Education & Science, 2015) Ugurlu, Bora; Yenisari, Esma; Karasulu, Bahadir; Ayan, Ozcan Zafer
    People who do not have expertise in the financial area may not see the relationship between the numerical and linguistic data. In our study, a knowledge discovery approach using Turkish natural language processing is recommended in order to respond to meaningful queries and classify them with high accuracy. Query corpus consists of randomly selected unique keywords. Quantitative evaluation is done in order to measure the classification performance. Experimental results indicate that our proposed approach is sufficiently consistent with and able to make categorical classifications correctly. The approach highlights the relationship between numerical and linguistic data obtained from Turkish financial market.
  • [ X ]
    Öğe
    A software approach to performance evaluation
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter provides four sections. The first section introduces the ViCamPEv software infrastructure (i.e., OpenCV, etc.), the CAVIAR video dataset, and its CVML (i.e., XML GT file format), and some other studies in the literature (e.g., MERL-PEP platform). In the second section, the architectural overview of the ViCamPEv software is presented and how the GT data is used to measure and evaluate the performance of given object D&T methods. Also, UML-based component diagram of software and the development platform of software are presented. In the third section, the testing conditions and testbed (i.e., software and hardware) are explained, and detailed via the screenshots of ViCamPEv’s GUI windows. In the fourth section, the system workflow is given via two parts of ViCamPEv: the Camera part and the Video part. The system is treated as automatic or semi-automatic test environment, which combines the input data with parameters of given methods. © The Author(s) 2013.
  • [ X ]
    Öğe
    An approach based on simulated annealing to optimize the performance of extraction of the flower region using mean-shift segmentation
    (Elsevier Science Bv, 2013) Karasulu, Bahadir
    Flower identification and recognition are tedious and difficult tasks even for humans. Image segmentation based on automatic flower extraction is an essential step for computer-aided flower image recognition and retrieval processes. Furthermore, there is a challenge for segmentation of the object(s) s) from natural complex background in color images. In this study, a novel performance optimization approach for image segmentation, i.e. simulated annealing-based mean-shift segmentation (SAMS), is proposed and implemented. It is based on the simulated annealing solution of quadratic assignment problem model treated as an image segmentation process using feature-based mean-shift (MS) clustering on color images. The proposed approach is designed to realize a global and unsupervised (i.e., fully automatic) segmentation. It is a modified and optimized version of Backprojection-based mean-shift segmentation (BackMS) method. In conducted segmentation experiments, the performance results of SAMS approach are compared with the ones of BackMS method. Comparison of overall performance results and statistical analysis (i.e., Wilcoxon signed rank median test) show that SAMS approach improves the performance of BackMS method. It is measured as 49.33% when using object bounding boxes and as 51.33% when using object pixel regions. (C) 2013 Elsevier B.V. All rights reserved.
  • [ X ]
    Öğe
    An Automatic Optic Disk Detection and Segmentation System using Multi-level Thresholding
    (Univ Suceava, Fac Electrical Eng, 2014) Karasulu, Bahadir
    Optic disk (OD) boundary localization is a substantial problem in ophthalmic image processing research area. In order to segment the region of OD, we developed an automatic system which involves a multi-level thresholding. The OD segmentation results of the system in terms of average precision, recall and accuracy for DRIVE database are 98.88%, 99.91%, 98.83%, for STARE database are 98.62%, 97.38%, 96.11%, and for DIARETDB1 database are 99.29%, 99.90%, 99.20%, respectively. The experimental results show that our system works properly on retinal image databases with diseased retinas, diabetic signs, and a large degree of quality variability.
  • [ X ]
    Öğe
    An Optimized Image Segmentation Approach Based on Boltzmann Machine
    (Taylor & Francis Inc, 2017) Karasulu, Bahadir
    Image segmentation with complex background is a tedious task. In our study, a convex spline is constructed based on Good Features to Track (GF2T) method's region-based salient feature (i.e., corner) set. For an optimized edge-based segmentation, an ellipse shape prior based on this convex spline is useful in edge regularization procedure with region-based features. This kind of optimization is achieved by Boltzmann machine (BM) to automatically form an elliptical foreground mask of the GrabCut method. We demonstrated our approach's usability through traveling salesman problem (TSP), thus, we consider that the TSP's valid tour's path solved by BM can be taken as an optimized convex spline for edge-based segmentation. In our experiments, proposed BM-based approach has the performance improvement of segmentation to stand-alone GF2T as 29.79% improvement based on bounding boxes evaluation and as 38.67% improvement based on the overlapping pixel regions for a quantitative evaluation via objective metrics.
  • [ X ]
    Öğe
    An overview on case-based reasoning and its synergetic applications with hybrid reasoning
    (Computers and Industrial Engineering, 2014) Ugurlu, Bora; Karasulu, Bahadir
    A variety of applications in quality management, health care, intelligent manufacturing and communication areas involves the systems based on case-based reasoning with an inference mechanism. The main aim of case-based reasoning is to solve a given problem defined in a specified experience area. A case-base or a case memory is used to make a decision for a problem. The base covers previously solved problem's solutions. For a hybridization, there are a lot of studies using synergism approach so as to combine different knowledge presentation models or reasoning methodologies. It fills possible knowledge gaps or existing lacks for individual systems. There are many advantages, such as simplifying the knowledge acquisition, obtaining an automatic explanation ability, building a cost-effective system and increasing the quality of problem solutions. In this study, we systematically reviewed remarkable studies based on case-based reasoning or its synergetic applications with hybrid reasoning in the literature. They are selected from the years between 1998 and 2013, and involve expert systems or similar mechanisms used in medical, industrial and business fields. The results are discussed.
  • [ X ]
    Öğe
    Anlamsal Boşluk Doldurulmasında Derin Öğrenme Yöntemlerinin Kullanılması Üzerine Bir İnceleme
    (Murat GÖK, 2024) Metin, İbrahim Ali; Karasulu, Bahadir
    Anlamsal boşluk kavramı, makineler aracılığıyla veriden elde edilen temel renk, doku ve şekil gibi özniteliklerle insan tarafından aynı veriye bakıldığında algı yoluyla tanımlanan kavramsal sonuçların farklılıklarından doğmaktadır. Bu boşluğun giderilmesi için çeşitli yol ve yöntemler literatürdeki çalışmalarda denenmiştir. Bu çalışmalarda, ontolojik sistemlerin geliştirildiği ve arama işlemlerinin de bu tarafa yoğunlaştırılmasının sağlandığı görülmektedir. Çalışmamızın ana amacı genel olarak Anlamsal Tabanlı Görüntüden Bilgi Elde Etme çalışmalarıyla İçerik Tabanlı Görüntüden Bilgi Elde Etme için gerçekleştirilen aramada ortaya çıkan Anlamsal Boşluk sorunun üstesinden gelinmesine dair yapılan çalışmaları ortaya koymaktır. Çalışmamızda literatürdeki çeşitli uluslararası dergi ve konferans yayınları incelenerek, kıyaslamalı özlü ve sistematik bir literatür taraması sunulmaktadır. Bilimsel tartışmaya da yer verilmektedir
  • [ X ]
    Öğe
    Arnavutça Konuşma Verilerini Kullanan Derin Öğrenme Tabanlı Duygu Durum Analizi ve Sınıflandırma
    (Murat GÖK, 2024) Karasulu, Bahadir; Avcı, Elif; Strazimiri, Tesnim; Cengiz, Betül
    Günümüzde konuşma veya ses verilerinden konuşmacının duygu durumunun analiz edilebildiği derin öğrenme tabanlı yazılımlar sayesinde etkileşimli sesli çağrı yanıtlama sistemleri oluşturulabilmektedir. Çalışmamızda, Arnavutça bir veri kümesi oluşturulmuştur. Ses verilerinin spektral ve duyusal açıdan analizi çeşitli derin öğrenme modelleri kullanılarak gerçekleştirilmiştir. Oluşturulan veri kümesi, dört farklı duygu sınıfını (öfkeli, mutlu, üzgün, şaşkın) içeren Arnavutça konuşma verilerini içermektedir. Sınıflandırma işleminde, evrişimli sinir ağı (ESA) modeli kullanılmıştır. Deneysel sonuçlara göre, Arnavutça duygu durumu sınıflandırma başarımı, alıcı işletim karakteristik (AİK) eğrisi altında kalan alan (EAKA) bazında; öfkeli sınıfı için 0.76, mutlu sınıfı için 1.00, üzgün sınıfı için 1.00 ve şaşkın sınıfı için 0.93 olarak elde edilmiştir. Çalışmaya dair bilimsel bulgular ve tartışmalar da sunulmuştur
  • [ X ]
    Öğe
    Automatic Segmentation of the Human Ear Using Active Contour and GrabCut Synergy Based on the Superpixel Cluster Regions
    (İstanbul Üniversitesi, 2021) Karasulu, Bahadir
    The ear region is a region of the human body region containing valuable biometric information that is subjected to a few physiological changes depending on the individual’s age. Manual, semi-automatic, or fully automatic segmentation of the ear region in various methods related to the use of the ear region in obtaining biometric information is an important area of research. In our study, we present an approach that applies superpixel cluster regions, active contour detection based on geodesic information, and foreground separation by graph cutting, to segregate the human ear region from the image by fully automatic segmentation from the background. Thanks to this approach in our study, the ear foreground mask is created programmatically and fully automatically from the ear image. In the experiments with the ear images data set, the reference ear mask marked by the expert was compared with the automatically created foreground mask. It has been obtained hHigh performance values were obtained, considering the similarity rates (i.e., intersection over union) based on the Jaccard index metric. Our approach has quite good performance values (in the range of 84% to 92%) for the images in this dataset. In our study, the success of the proposed synergistic approach is demonstrated both qualitatively and quantitatively with experimental results
  • [ X ]
    Öğe
    Clustering Methods Comparison for Optimization of Adaptive Neural Fuzzy Inference System
    (IEEE, 2022) Fidan, Sertug; Karasulu, Bahadir
    Different methods have been developed to optimize the Adaptive Neural Fuzzy Inference System, which is used in many fields due to its flexible structure and trainability. Within the scope of this study, three different models were produced using two different datasets, using only the first clustering method, only the second clustering method, and both the first and second clustering methods. In this study, the Fuzzy C-Mean Clustering algorithm, which is one of the most efficient methods used to reduce the number of rules in the rule base of the hybrid intelligent system is compared with the Highly Connected Subgraphs algorithm. The models were compared over the square root of the mean square error, the number of nodes, the number of fuzzy rules, and the mean training time. As a result of the study, the second clustering method formed the most efficient result in terms of error rate with 0.084 and 0,008. It has been observed that the average training time of this method is approximately 31 times longer than the first clustering method mentioned above, and approximately 52 times longer than the model in which the first and second clustering methods are used together. In this study, it has been seen that the first clustering method is more successful in reducing the rule base by optimizing the second method by determining more suitable cluster centers. Based on the experimental results obtained in our study, these two different clustering methods were compared over three different models. Discussion and scientific results are included in our study.
  • [ X ]
    Öğe
    Conclusion
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter concludes the book, also it involves a summary of the study. Therefore, it declares some contributions of given study. In addition, this chapter suggests to readers or researchers in computer vision and multimedia research area that the ViCamPEv software is useful for image and video processing, multimedia content, and information retrieval as well. © The Author(s) 2013.
  • [ X ]
    Öğe
    Derin Öğrenme Modellerini Kullanarak İnsan Retinasının Optik Koherans Tomografi Görüntülerinden Hastalık Tespiti
    (Murat GÖK, 2022) Metin, Batuhan; Karasulu, Bahadir
    Bireylerin yaşamını olumsuz etkileyen aynı zamanda bireylerin yaşlarının ilerlemesi ile yaşamın kaçınılmazlarından olan en önemli sorunlardan birisi de retina hastalıkları sebebi ile meydana gelen görme bozukluklarıdır. Bu hastalıkların oluşmasının önüne geçmek için erken dönemlerde teşhis etme sayesinde yaşamın olumsuz etkilenmesinde ve sonraki evresi olan görme kaybı riskini en aza indirmek için çok önemlidir. Gelişen teknolojik yöntemler ile doğru orantılı olarak kullanımı yaygınlaşan makine öğrenmesi ve derin öğrenme yöntemleri Optik Koherans Tomografisi (OKT) görüntüleme yöntemi üzerinde çalışmayı yaygınlaştırmıştır. Bu çalışmada halkın kullanımına açık OKT veri kümesi üzerinden deneyler gerçekleştirilmiştir. Yüksek sınıflandırma performansları göz önüne alınarak Evrişimli Sinir Ağı (ESA) tabanlı ResNet50 ve MobileNetV2 modelleri çalışmamızda kullanılmıştır. Oluşturulan derin öğrenme tabanlı yapılar çalışmamızda gerçekleştirilen deneylerde çeşitli retina hastalıklarının sınıflandırılmasında test edilmiştir. Deneylerde farklı parametreler oluşturulan modeller üzerinde girdi olarak verilerek sınıflandırma başarımındaki doğruluk ölçümleri gerçekleştirilmiştir. Yapılan testlerin sonucunda, her iki model de dikkate alınarak makro ortalama doğruluk değerleri olarak yaklaşık %81 ile %94 aralığında bir başarım elde edilmiştir. Bu test sonuçlarına göre; deneylerde kullanılan ResNet50 ve MobileNetV2 modelleri birlikte dikkate alındığında ortalama F1 skoru; CNV için 0,75, Drusen için 0,86, DME için 0,90 ve normal yapıdaki retinalar için ise 0,96 olarak elde edilmiştir. Elde edilen sonuçlar ile literatürdeki çalışmalar karşılaştırıldığında yüksek doğrulukta başarım elde edildiği görülmüştür. Çalışmada sonuçlara ilişkin tartışma ve bilimsel bulgulara da yer verilmiştir
  • [ X ]
    Öğe
    Introduction
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter comprises three sections. The first section represents the scope of this book. The second section introduces the related works on the issue of moving object detection and tracking (D&T) in videos. Also, the main objective for moving object D&T and main scenarios for visual surveillance applications is given in this chapter. The moving object D&T methods in video processing are categorized in some ways where their respective aspects are taken as the basis of the D&T process. In the second section, the basis of performance and evaluation of moving object D&T process is declared, such as with-ground-truth and without-ground-truth evaluation. In addition, commonly used video datasets, their tools, and some systems for object D&T are briefly introduced. This chapter ends with a third section focused on the main contribution of the study given in this book. © The Author(s) 2013.
  • [ X ]
    Öğe
    Moving object detection and tracking in videos
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter provides four sections. The first section introduces the moving object D&T infrastructure and basis of some methods for object detection and tracking (D&T) in videos. In object D&T applications, there is manual or automatic D&T process. Also, the image features, such as color, shape, texture, contours, and motion can be used to track the moving object(s) in videos. The detailed information for moving object detection and well-known trackers are presented in this section as well. In second section, the background subtraction (BS) method and its applications are given in details. The third section declares the details for Mean-shift (MS), Mean-shift filtering (MSF), and continuously adaptive Mean-shift (CMS or CAMShift) methods and their applications. In fourth section, the details for the optical flow (OF), the corner detection through feature points, and OF-based trackers are given in details. © The Author(s) 2013.
  • [ X ]
    Öğe
    PARALLEL COMPUTING IN FUZZY DECISION MAKING SYSTEMS
    (Pamukkale Univ, 2013) Balli, Serkan; Karasulu, Bahadir
    In some decision problems, the numbers of alternatives, criteria and decision-makers are very high. Therefore the calculation process becomes more difficult, time consuming and complex, To achieve these complex tasks in a shorter time, popular technologies such us parallel computing are available to use. In this study, design and implementation of parallel computation in a fuzzy decision making system which consists of the methods: Fuzzy Analytic Hierarchy Process (FAHP) and TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) was investigated. Parallel computing was carried out in FAHP phase that is most intensive phase of calculation. Proposed method was tested in homogeneous and heterogeneous computers separately and results were discussed.
  • [ X ]
    Öğe
    Performance evaluation software: Moving object detection and tracking in videos
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    [No abstract available]
  • [ X ]
    Öğe
    Performance measures and evaluation
    (Springer, 2013) Karasulu, Bahadir; Korukoglu, Serdar
    This chapter provides two sections. The first section introduces our performance evaluation methodology. The second section is separated into three subsections, which are frame-based detection measures, measures based on object matching criteria, and object tracking-based measures, respectively. In our study, some performance measures are used to evaluate object D&T method’s performance. In this manner, the measures for frame-based detection are object count accuracy, pixel-based precision, pixel-based recall, pixel-based F1 measure, area-based precision, area-based recall, area-thresholded precision, area-thresholded recall, and average fragmentation, respectively. Also, the measures based on object matching criteria are sequence frame detection accuracy, and position-based, size-based, and orientation-based measures, respectively. In addition, the measures based on object tracking are the object count-based measure, temporal measure, and sequence tracking detection accuracy, respectively. All of the above-mentioned measures are given via mathematical background in related subsections of this book. © The Author(s) 2013.
  • [ X ]
    Öğe
    Ses Özniteliklerini Kullanan Ses Duygu Durum Sınıflandırma İçin Derin Öğrenme Tabanlı Bir Yazılımsal Araç
    (Murat GÖK, 2021) Kıvrak, Emir Ali; Karasulu, Bahadir; Sözbir, Can; Türkay, Atakan
    Ses duygu durum analizi için kullanıcı grafik arayüzü yardımıyla ses verilerini kullanarak ses duygu durumları herhangi bir kaynak kodu satırı yazmadan sınıflandıran derin öğrenme mimari modellerini oluşturan bir yazılımsal araç çalışmamızda tasarlanmıştır. Veri kümelerinin elde edilmesi, ses verilerine yönelik ses özniteliklerinin elde edilmesi, mimarinin oluşturulması ve derin öğrenme modelinin istenilen sinir ağı katmanları ve üstün parametreler ile modelin eğitilmesi sağlanmıştır. Model eğitilirken, eğitim değerlerinin gerçek zamanlı izlenmesi yazılımsal araç ile yapılabilmektedir. Çalışma boyunca, ilgili adımlar hem salt kaynak kodu düzenleme hem de yazılımsal araç kullanılarak gerçekleştirilmiştir. Kod düzenleme tabanlı melez model, mimarisinde uzun kısa süreli bellek ve evrişimli sinir ağları kullanılarak oluşturulmuş, %81,49 doğruluk oranına ulaşmıştır. Ayrıca, herhangi bir kodlama müdahalesi olmaksızın grafik yazılımsal araç tabanlı tekil model, mimarisinde evrişimli sinir ağı ile oluşturulmuştur. Böylece %75,76 doğruluk oranına ulaşmıştır. Yazılımsal aracın geliştirilmesindeki ana motivasyon, farklı ses duygu durumları sınıflandırmak için kullanılabilecek potansiyel bir derin öğrenme mimari modeli oluşturmaktır. Deneysel sonuçlar, yazılımsal aracın yüksek doğrulukla sınıflandırmayı oldukça başarılı bir şekilde gerçekleştirdiğini kanıtlamaktadır. Elde edilen sonuçlara dair tartışmaya da çalışmamızda yer verilmiştir