Araştırma Çıktıları | WoS | Scopus | TR-Dizin | PubMed
Permanent URI for this communityhttps://hdl.handle.net/20.500.14719/1741
Browse
10 results
Search Results
Publication Metadata only ATMOSPHERIC TURBULENCE MITIGATION USING FEATURE BASED OPTICAL FLOW(IEEE, 2014) Caliskan, Tufan; Arica, Nafiz; Deniz Harp Okulu Komutanligi; Bahcesehir UniversityIn this paper, we propose a fast and effective feature based registration technique in removing the distortions caused by atmospheric turbulence. We utilize optical flow method and combine the advantages of previous approaches based on space-invariant deconvolution and lucky frame idea. After an optical flow based registration of degraded image sequence, a patch-wise multi-frame reconstruction technique is applied to fuse the registered images. Blind-deconvolution technique is implemented to deblur the fused image to obtain a single high quality image. The experiments performed on common datasets show that the proposed method produces higher quality images more efficiently than the available methods.Publication Metadata only Automatically Inserting Ads into Images(IEEE, 2018) Citak, Erol; Eseoglu, Mustafa Furkan; Celik, Ibrahim Omer; Disanli, Onur; Kutluk, Sezer; Arica, Nafiz; Huawei Technologies; Bahcesehir UniversityIn this study a new method is proposed for inserting advertisement visuals into images automatically and without disturbing the image content. In this method important areas are determined using deep learning based object, face and text detection, edge and saliency maps are obtained, and these information are used for the identification of the best location for inserting the advertisement visual. In order to select the best available advertisement visual from an advertisement pool shape and color features are utilized.Publication Metadata only Facial Emotion Classification Using Deep Embedding with Triplet Loss Function(IEEE, 2017) Bircanoglu, Cenk; Arica, Nafiz; Bahcesehir UniversityIn this paper, a deep embedding method using triplet loss function is proposed for classification of the emotions in face images. The originality of proposed method lies in the loss function, different from the other deep learning-based facial emotion classification approaches. The input face images are embedded into a lower dimensional feature space using a multi layer convolutional neural network. In this embedding process the loss function is calculated by taking triple samples in the training dataset. For each sample in batches, two samples are selected in such a way that one of them is from the same class and the other one from the different class. The loss function aims to close the samples belonging to the same class to each other in the Euclidean space and to move away from the samples in the different class. In the performance analysis of proposed method, two popular deep architectures, namely AlexNet and VGG are used on two datasets called CIFE and GaMo. The experiments show that the proposed method outperforms the other popular loss functions and the available studies on this particular datasets.Publication Metadata only Auto-encoder based Cognitive Analysis of Questionnary(IEEE, 2018) Akay, Simge; Korkmaz, Hande; Arica, Nafiz; Bahcesehir University; Bahcesehir UniversityIn this study we propose a new method for Likert scale questionnary data analysis using auto-encoders. The proposed method extracts the patterns, which maximally activate the neurons of the hidden layer in the auto-encoder, trained by the questionnary data. These patterns are thought to be cognitive patterns that are influenced by participants filling the questionnary. In the experiments, we employ a questionnary designed to measure the confidence level of a blog author on the web. The cognitive patterns obtained in the auto-encoder are considered as the components that form the general approaches of the participants. In two cognitive patterns drawn from the questionnary, it is observed that the blogger's expertise, integrity, benevolance are evaluated in decreasing or increasing order by the participants. It has also been observed that the proposed method can be used to correct the unintentional mistakes in questionnary answers.Publication Metadata only Face Frontalization Enhanced by Deep Learning(IEEE, 2017) Celik, Anil; Arica, Nafiz; Bahcesehir UniversityIn this study, a new approach based on 3-D models and deep learning to frontalize face images is proposed. Specifically designed for facial expression analysis, the proposed approach aims to reduce possible negative effects that a posed face image can generate, by normalizing the face region. In the first phase, the face image is semi-frontalized, with a pre-established 3-D reference model based approach. Then, missing regions on semi-frontalized images due to geometric transformation are reconstructed with the help of a denoising stacked autoencoder network. In this phase, missing regions created by line of sight are learned, with a deep architecture, using numerous images. When examined, it can be said that, faces acquired with the proposed approach, are objectively better than the faces acquired with a deep learning or 3-D based method alone. Therefore, it is assumed that the proposed approach can be used in face based computer vision methods as a beneficial pre-processing step.Publication Metadata only AN HIERARCHICAL APPROACH FOR HUMAN COMPUTER INTERACTION USING EYELID MOVEMENTS(IEEE, 2014) Celik, Anil; Arica, Nafiz; Bahcesehir University; Bahcesehir UniversityThis work proposes a method to achieve real-time Human-Computer interaction with the movements of the eyelids in low-resolution video. Classification of left and right eye states as closed or open are performed by an hierarchical approach of tracking by detection. After the initial detection of the face area, an efficient face tracking algorithm is used to reduce the search space for detection of eye region. By seperating the eye region into two overlapping pieces, left and right eyes are detected and classified as closed or open. The proposed method is robust against mimics and multiple faces in the frame, while being unaffected by the negative effects of aliasing and resizing. Thus, people who suffer from medical conditions limiting their physical movement capabilities can interact with computers with their eyelid movements.Publication Metadata only Facial Action Unit Detection Using Deep Neural Networks in Videos(IEEE, 2018) Akay, Simge; Arica, Nafiz; Bahcesehir UniversityThe detection of facial action unit is one of the most important sources for describing facial expressions. Some reasons such as facial expressions in the video, changes in the environment, different posed face images make it more difficult to detect facial action unit. In this work, a new approach, which is based on deep neural network, is recommended for facial action unit detection. The recommended approach uses three different types of classifier which are frame based, segment based and transition based while detecting the facial action unit. In classification stage, Motion History Images are given as input to the deep network in addition to the pixel values of images. Finally, results of three different classifiers for each frame is combined by using the Support Vector Machine. In the experiments which are done by using CK+ database, it is observed that the recommended approach gives us more successful results of detection than the recent studies in the literature.Publication Metadata only Size Variant Landmark Patches for Facial Action Unit Detection(IEEE, 2016) Cakir, Duygu; Arica, Nafiz; Chakrabarti, S; Saha, HN; Bahcesehir UniversityFacial Action Coding System (FACS) provides the description of all visual facial changes in terms of Action Units (AU), which express the movements of facial muscles. In this study we introduce a sparse learning method for AU detection based on facial landmark patches. The proposed method learns the active landmarks for each AU while finding their best representative patch sizes in a unified framework. The experiments performed on the publicly available CK+ dataset show that the proposed approach outperforms the state-of-the-art studies by gaining the highest F-1 scores for almost all of the lower face Action Units. The experiments also show that the subtle muscle movements belonging to the upper face require smaller landmark patches while the lower face AUs are detected better in larger patches.Publication Metadata only A Case Study on Transfer Learning in Convolutional Neural Networks(IEEE, 2018) Gurkaynak, Cahit Deniz; Arica, Nafiz; Bahcesehir UniversityIn this work, a case study is performed on transfer learning approach in convolutional neural networks. Transfer learning parameters are examined on AlexNet, VGGNet and ResNet architectures for marine vessel classification task on MARVEL dataset. The results confirmed that transferring the parameter values of the first layers and fine-tuning the other layers, whose weights are initialized from pre-trained weights, performs better than training network from scratch. It's also observed that preprocessing and regularization improves overall scores significantly.Publication Metadata only A Comparison of Activation Functions in Artificial Neural Networks(IEEE, 2018) Bircanoglu, Cenk; Arica, Nafiz; Bahcesehir UniversityIn this study, the effects of Activation Functions (AF) in Artificial Neural Network (ANN) on regression and classification performance are compared. In comparisons, success rates in test data and duration of training are evaluated for both problems. A total of 11 AF functions, 10 AF commonly used in the literature and Square function proposed in this study, are compared using 7 different datasets, 2 for regression and 5 for classification. 3 different ANN architectures, which are considered to be the most appropriate for each dataset are employed in the experiments. As a result of totally 231 different training procedures, the effects of Afs are examined for different datasets and architectures. Similarly, the effects of AF on training time are shown for different datasets. In the experiments it is shown that ReLU is the most succesfull AF in general purposes. In addition to ReLU, Square function gives the better results in image datasets.