Browsing by Author "Nar, Fatih"

Now showing 1 - 5 of 5

Citation Count: Saran, Nurdan Ayşe; Saran, Murat; Nar, Fatih (2021). "Distribution-preserving data augmentation", Peerj Computer Science.
Distribution-preserving data augmentation
(2021) Saran, Nurdan Ayşe; Saran, Murat; Nar, Fatih; 20868; 17753
In the last decade, deep learning has been applied in a wide range of problems with tremendous success. This success mainly comes from large data availability, increased computational power, and theoretical improvements in the training phase. As the dataset grows, the real world is better represented, making it possible to develop a model that can generalize. However, creating a labeled dataset is expensive, time-consuming, and sometimes not likely in some domains if not challenging. Therefore, researchers proposed data augmentation methods to increase dataset size and variety by creating variations of the existing data. For image data, variations can be obtained by applying color or spatial transformations, only one or a combination. Such color transformations perform some linear or nonlinear operations in the entire image or in the patches to create variations of the original image. The current color-based augmentation methods are usually based on image processing methods that apply color transformations such as equalizing, solarizing, and posterizing. Nevertheless, these color-based data augmentation methods do not guarantee to create plausible variations of the image. This paper proposes a novel distribution-preserving data augmentation method that creates plausible image variations by shifting pixel colors to another point in the image color distribution. We achieved this by defining a regularized density decreasing direction to create paths from the original pixels' color to the distribution tails. The proposed method provides superior performance compared to existing data augmentation methods which is shown using a transfer learning scenario on the UC Merced Land-use, Intel Image Classification, and Oxford-IIIT Pet datasets for classification and segmentation tasks.
Citation Count: Özgür, Atilla; Saran, Ayşe Nurdan; Nar, Fatih, "Parallelization of sparsity-driven change detection method", 2017 25th Signal Processing And Communications Applications Conference (SIU), (2017).
Parallelization of sparsity-driven change detection method
(IEEE, 2017) Özgür, Atilla; Saran, Ayşe Nurdan; Nar, Fatih; 20868
In this study, Sparsity-driven Change Detection (SDCD) method, which has been proposed for detecting changes in multitemporal synthetic aperture radar (SAR) images, is parallelized to reduce the execution time. Parallelization of the SDCD is realized using OpenMP on CPU and CUDA on GPU. Execution speed of the parallelized SDCD is shown on real-world SAR images. Our experimental results show that the computation time of the parallel implementation brings significant speed-ups.
Citation Count: Saran, Murat; Nar, Fatih; Saran, Ayse Nurdan (2021). "Perlin random erasing for data augmentation", SIU 2021 - 29th IEEE Conference on Signal Processing and Communications Applications, Proceedings, 29th IEEE Conference on Signal Processing and Communications Applications, SIU 2021Virtual, Istanbul9 June 2021through 11 June 2021.
Perlin random erasing for data augmentation
(2021) Saran, Murat; Nar, Fatih; Saran, Ayse Nurdan; 17753; 20868
In the last decade, Deep Learning is applied in a wide range of problems with tremendous success. Large data, increased computational resources, and theoretical improvements are main reasons for this success. As the dataset grows, the realworld is better represented, allows developing a model that can generalize. However, creating a labeled dataset is expensive, timeconsuming, or sometimes even challenging. Therefore, researchers proposed data augmentation methods to increase dataset size by creating variations of the existing data. This study proposes an extension to Random Erasing data augmentation method by introducing smoothness. The proposed method provides better performance compared to Random Erasing data augmentation method, which is shown using a transfer learning scenario on the UC Merced Land-use image dataset.
Citation Count: Nar, F., Özgür,A., Saran, A.N. (2016). Sparsity-driven change detection in multitemporal sar images. IEEE Geoscience And Remote Sensing Letters, 13(7), 1032-1036. http://dx.doi.org/10.1109/LGRS.2016.2562032
Sparsity-driven change detection in multitemporal sar images
(IEE-INST Electrical Electronics Engineers Inc., 2016) Nar, Fatih; Özgür, Atilla; Saran, Ayşe Nurdan; 252953; 20868
In this letter, a method for detecting changes in multitemporal synthetic aperture radar (SAR) images by minimizing a novel cost function is proposed. This cost function is constructed with log-ratio-based data fidelity terms and an l(1)-norm-based total variation (TV) regularization term. Log-ratio terms model the changes between the two SAR images where the TV regularization term imposes smoothness on these changes in a sparse manner such that fine details are extracted while effects like speckle noise are reduced. The proposed method, sparsity-driven change detection (SDCD), employs accurate approximation techniques for the minimization of the cost function since data fidelity terms are not convex and the employed l(1)-norm TV regularization term is not differentiable. The performance of the SDCD is shown on real-world SAR images obtained from various SAR sensors.
Citation Count: Saran, Ayşe Nurdan; Saran, Murat; Nar, Fatih, "Vessel segmentation in MRI using a variational image subtraction approach", Turkish Journal of Electrical Engineering and Computer Sciences, Vol. 22, No. 2, pp. 499-516, (2014).
Vessel segmentation in MRI using a variational image subtraction approach
(2014) Saran, Ayşe Nurdan; Nar, Fatih; Saran, Murat; 20868; 17753
Vessel segmentation is important for many clinical applications, such as the diagnosis of vascular diseases, the planning of surgery, or the monitoring of the progress of disease. Although various approaches have been proposed to segment vessel structures from 3-dimensional medical images, to the best of our knowledge, there has been no known technique that uses magnetic resonance imaging (MRI) as prior information within the vessel segmentation of magnetic resonance angiography (MRA) or magnetic resonance venography (MRV) images. In this study, we propose a novel method that uses MRI images as an atlas, assuming that the patient has an MRI image in addition to MRA/MRV images. The proposed approach intends to increase vessel segmentation accuracy by using the available MRI image as prior information. We use a rigid mutual information registration of the MRA/MRV to the MRI, which provides subvoxel accurate multimodal image registration. On the other hand, vessel segmentation methods tend to mostly suffer from imaging artifacts, such as Rician noise, radio frequency (RF) inhomogeneity, or partial volume effects that are generated by imaging devices. Therefore, this proposed method aims to extract all of the vascular structures from MRA/MRI or MRV/MRI pairs at the same time, while minimizing the combined effects of noise and RF inhomogeneity. Our method is validated both quantitatively and visually using BrainWeb phantom images and clinical MRI, MRA, and MRV images. Comparison and observer studies are also realized using the BrainWeb database and clinical images. The computation time is markedly reduced by developing a parallel implementation using the Nvidia compute unified device architecture and OpenMP frameworks in order to allow the use of the method in clinical settings.