15 results
Search Results
Now showing 1 - 10 of 15
Article Citation - WoS: 2Citation - Scopus: 3A Hybrid Approach for Semantic Image Annotation(Ieee-inst Electrical Electronics Engineers inc, 2021) Sezen, Arda; Turhan, Cigdem; Sengul, GokhanIn this study, a framework that generates natural language descriptions of images within a controlled environment is proposed. Previous work on neural networks mostly focused on choosing the right labels and/or increasing the number of related labels to depict an image. However, creating a textual description of an image is a completely different phenomenon, structurally, syntactically, and semantically. The proposed semantic image annotation framework presents a novel combination of deep learning models and aligned annotation results derived from the instances of the ontology classes to generate sentential descriptions of images. Our hybrid approach benefits from the unique combination of deep learning and semantic web technologies. We detect objects from unlabeled sports images using a deep learning model based on a residual network and a feature pyramid network, with the focal loss technique to obtain predictions with high probability. The proposed framework not only produces probabilistically labeled images, but also the contextual results obtained from a knowledge base exploiting the relationship between the objects. The framework's object detection and prediction performances are tested with two datasets where the first one includes individual instances of images containing everyday scenes of common objects and the second custom dataset contains sports images collected from the web. Moreover, a sample image set is created to obtain annotation result data by applying all framework layers. Experimental results show that the framework is effective in this controlled environment and can be used with other applications via web services within the supported sports domain.Article Citation - WoS: 9Citation - Scopus: 9Benchmarking Classification Models for Cell Viability on Novel Cancer Image Datasets(Bentham Science Publ Ltd, 2019) Ozkan, Akin; Isgor, Sultan Belgin; Sengul, Gokhan; Isgor, Yasemin GulgunBackground: Dye-exclusion based cell viability analysis has been broadly used in cell biology including anticancer drug discovery studies. Viability analysis refers to the whole decision making process for the distinction of dead cells from live ones. Basically, cell culture samples are dyed with a special stain called trypan blue, so that the dead cells are selectively colored to darkish. This distinction provides critical information that may be used to expose influences of the studied drug on considering cell culture including cancer. Examiner's experience and tiredness substantially affect the consistency throughout the manual observation of cell viability. The unsteady results of cell viability may end up with biased experimental results accordingly. Therefore, a machine learning based automated decision-making procedure is inevitably needed to improve consistency of the cell viability analysis. Objective: In this study, we investigate various combinations of classifiers and feature extractors (i.e. classification models) to maximize the performance of computer vision-based viability analysis. Method: The classification models are tested on novel hemocytometer image datasets which contain two types of cancer cell images, namely, caucasian promyelocytic leukemia (HL60), and chronic myelogenous leukemia (K562). Results: From the experimental results, k-Nearest Neighbor (KNN) and Random Forest (RF) by combining Local Phase Quantization (LPQ) achieve the lowest misclassification rates that are 0.031 and 0.082, respectively. Conclusion: The experimental results show that KNN and RF with LPQ can be powerful alternatives to the conventional manual cell viability analysis. Also, the collected datasets are released from the "biochem.atilim.edu.tr/datasets/ " web address publically to academic studies.Article Crack Detection on Asphalt Runway Using Unmanned Aerial Vehicle Data with Non-Crack Object Removal and Deep Learning Methods(Pontificia Univ Catolica Chile, Escuela Construccion Civil, 2025) Tapkin, Serkan; Tercan, Emre; Bostan, Atila; Sengul, GokhanUnmanned aerial vehicles are extensively utilized for image acquisition in a cheap, fast, and effective way. In this study, an automatic crack detection method with non-crack object removal and deep learning-based approaches are developed and tested on images captured by unmanned aerial vehicle. The motivation of this study is to detect either a crack exists or not in the asphalt-runway. The novelty of this study lies in integrating a non-crack artifact removal process with six classical edge detectors and comparing the resulting performance with four lightweight CNN models on the same UAV-acquired runway image dataset, enabling a unified evaluation of classical and learning-based approaches. For deep learning-based approach, four lightweight CNN models, namely GoogleNet, SqueezeNet, MobileNetv2, and ShuffleNet, are trained and the best accuracy of %87.9 is obtained whenever GoogleNet model is used. For the non-crack object removal approach, exclusion of non-crack objects from the images is the first step, where crack-detection which makes use of edge-detection techniques is the latter. In the study, Sobel, Prewitt, Canny, Laplacian of Gaussian, Roberts and Zero Cross edge detection algorithms are examined and their success rates in detecting cracks are comparatively presented. With sensitivity=0.981, specificity=0.744, accuracy=0.917, precision=0.912 and F-score=0.945 values Canny algorithm performs significantly better than others in detecting the cracks. This study provides enough evidence for the practicability of automated crack detection on unprocessed digital photographs by the results of the study conducted on asphalt runway.Article Citation - WoS: 8Citation - Scopus: 13White Blood Cells Classifications by SURF Image Matching, PCA and Dendrogram(Allied Acad, 2015) Nazlibilek, Sedat; Karacor, Deniz; Erturk, Korhan Levent; Sengul, Gokhan; Ercan, Tuncay; Aliew, Fuad; Department of Mechatronics Engineering; Information Systems Engineering; Computer EngineeringDetermination and classification of white blood cells are very important for diagnosing many diseases. The number of white blood cells and morphological changes or blasts of them provide valuable information for the positive results of the diseases such as Acute Lymphocytic Leucomia (ALL). Recognition and classification of white cells as basophils, lymphocytes, neutrophils, monocytes and eosinophils also give additional information for the diagnosis of many diseases. We are developing an automatic process for counting, size determination and classification of white blood cells. In this paper, we give the results of the classification process for which we experienced a study with hundreds of images of white blood cells. This process will help to diagnose especially ALL disease in a fast and automatic way. Three methods are used for classification of five types of white blood cells. The first one is a new algorithm utilizing image matching for classification that is called the Speed-Up Robust Feature detector (SURF). The second one is the PCA that gives the advantage of dimension reduction. The third is the classification tree called dendrogram following the PCA. Satisfactory results are obtained by two techniques.Article Citation - WoS: 49Citation - Scopus: 69Deep Learning Based Fall Detection Using Smartwatches for Healthcare Applications(Elsevier Sci Ltd, 2022) Sengul, Gokhan; Karakaya, Murat; Misra, Sanjay; Abayomi-Alli, Olusola O.; Damasevicius, RobertasWe implement a smart watch-based system to predict fall detection. We differentiate fall detection from four common daily activities: sitting, squatting, running, and walking. Moreover, we separate falling into falling from a chair and falling from a standing position. We develop a mobile application that collects the acceleration and gyroscope sensor data and transfers them to the cloud. In the cloud, we implement a deep learning algorithm to classify the activity according to the given classes. To increase the number of data samples available for training, we use the Bica cubic Hermite interpolation, which allows us to improve the accuracy of the neural network. The 38 statistical data features were calculated using the rolling update approach and used as input to the classifier. For activity classification, we have adopted the bi-directional long short-term memory (BiLSTM) neural network. The results demonstrate that our system can detect falling with an accuracy of 99.59% (using leave-one-activityout cross-validation) and 97.35% (using leave-one-subject-out cross-validation) considering all activities. When considering only binary classification (falling vs. all other activities), perfect accuracy is achieved.Article Citation - WoS: 18Citation - Scopus: 24Fusion of Smartphone Sensor Data for Classification of Daily User Activities(Springer, 2021) Sengul, Gokhan; Ozcelik, Erol; Misra, Sanjay; Damasevicius, Robertas; Maskeliunas, RytisNew mobile applications need to estimate user activities by using sensor data provided by smart wearable devices and deliver context-aware solutions to users living in smart environments. We propose a novel hybrid data fusion method to estimate three types of daily user activities (being in a meeting, walking, and driving with a motorized vehicle) using the accelerometer and gyroscope data acquired from a smart watch using a mobile phone. The approach is based on the matrix time series method for feature fusion, and the modified Better-than-the-Best Fusion (BB-Fus) method with a stochastic gradient descent algorithm for construction of optimal decision trees for classification. For the estimation of user activities, we adopted a statistical pattern recognition approach and used the k-Nearest Neighbor (kNN) and Support Vector Machine (SVM) classifiers. We acquired and used our own dataset of 354 min of data from 20 subjects for this study. We report a classification performance of 98.32 % for SVM and 97.42 % for kNN.Article Citation - WoS: 4Citation - Scopus: 7The Integrated Usage of Lbp and Hog Transformations and Machine Learning Algorithms for Age Range Prediction From Facial Images(Univ Osijek, Tech Fac, 2018) Khalifa, Tariq; Sengul, GokhanAge prediction is an active study field that can be used in many computer vision problems due to its importance and effectiveness. In this paper, we present extensive experiments and provide an efficient and accurate approach for age range prediction of people from facial images. First, we apply image resizing to unify all images' size, and Histogram Equalization technique to reduce the illumination effects on all facial images taken from FG-NET and UTD aging databases. Second, Histogram of Oriented Gradient (HOG) and Local Binary Pattern (LBP) are used to extract the features of these images, and then we combined both HOG and LBP features in order to attain better prediction. Finally, Support Vector Machine (SVM) and k-Nearest Neighbour (k-NN) are used for the classification processes. In addition, k-fold, Leave-One-Out (LOO) and Confusion Matrix (CM) are used to evaluate the performance of proposed methods. The extensive and intensified experiments show that combining HOG and LBP features improved the age range predicting performance up to 99.87%.Article Citation - WoS: 10Citation - Scopus: 31An Improved Random Bit-Stuffing Technique With a Modified Rsa Algorithm for Resisting Attacks in Information Security (rbmrsa)(Cairo Univ, Fac Computers & information, 2022) Mojisola, Falowo O.; Misra, Sanjay; Febisola, C. Falayi; Abayomi-Alli, Olusola; Sengul, GokhanThe recent innovations in network application and the internet have made data and network security the major role in data communication system development. Cryptography is one of the outstanding and powerful tools for ensuring data and network security. In cryptography, randomization of encrypted data increases the security level as well as the Computational Complexity of cryptographic algorithms involved. This research study provides encryption algorithms that bring confidentiality and integrity based on two algorithms. The encryption algorithms include a well-known RSA algorithm (1024 key length) with an enhanced bit insertion algorithm to enhance the security of RSA against different attacks. The security classical RSA has depreciated irrespective of the size of the key length due to the development in computing technology and hacking system. Due to these lapses, we have tried to improve on the contribution of the paper by enhancing the security of RSA against different attacks and also increasing diffusion degree without increasing the key length. The security analysis of the study was compared with classical RSA of 1024 key length using mathematical evaluation proofs, the experimental results generated were compared with classical RSA of 1024 key length using avalanche effect in (%) and computational complexity as performance evaluation metrics. The results show that RBMRSA is better than classical RSA in terms of security but at the cost of execution time. (C) 2022 THE AUTHORS. Published by Elsevier B.V. on behalf of Faculty of Computers and Information, Cairo University.Article Citation - WoS: 19Citation - Scopus: 24Gender Detection Using 3d Anthropometric Measurements by Kinect(Polska Akad Nauk, Polish Acad Sciences, 2018) Camalan, Seda; Sengul, Gokhan; Misra, Sanjay; Maskeliunas, Rytis; Damasevicius, RobertasAutomatic gender detection is a process of determining the gender of a human according to the characteristic properties that represent the masculine and feminine attributes of a subject. Automatic gender detection is used in many areas such as customer behaviour analysis, robust security system construction, resource management, human-computer interaction, video games, mobile applications, neuro-marketing etc., in which manual gender detection may be not feasible. In this study, we have developed a fully automatic system that uses the 3D anthropometric measurements of human subjects for gender detection. A Kinect 3D camera was used to recognize the human posture, and body metrics are used as features for classification. To classify the gender, KNN, SVM classifiers and Neural Network were used with the parameters. A unique dataset gathered from 29 female and 31 male (a total of 60 people) participants was used in the experiment and the Leave One Out method was used as the cross-validation approach. The maximum accuracy achieved is 96.77% for SVM with an MLP kernel function.Article Citation - WoS: 1Citation - Scopus: 3Three-Dimensional Visualization With Large Data Sets: a Simulation of Spreading Cortical Depression in Human Brain(Hindawi Publishing Corporation, 2012) Erturk, Korhan Levent; Sengul, GokhanWe developed 3D simulation software of human organs/tissues; we developed a database to store the related data, a data management system to manage the created data, and a metadata system for the management of data. This approach provides two benefits: first of all the developed system does not require to keep the patient's/subject's medical images on the system, providing less memory usage. Besides the system also provides 3D simulation and modification options, which will help clinicians to use necessary tools for visualization and modification operations. The developed system is tested in a case study, in which a 3D human brain model is created and simulated from 2D MRI images of a human brain, and we extended the 3D model to include the spreading cortical depression (SCD) wave front, which is an electrical phoneme that is believed to cause the migraine.
