Can Demirkesen

Region based target detection approach for synthetic aperture radar images and its parallel implementation

Automatic target detection (ATD) methods for synthetic aperture radar (SAR) imagery are sensitive to image resolution, target size, clutter complexity, and speckle noise level. However, a robust ATD method needs to be less sensitive to the above factors. In this study, a constant false alarm rate (CFAR) based method is proposed which can perform target detection independent of image resolution and target size even in heterogeneous background clutter. The proposed method is computationally efficient since clutter statistics are calculated only for candidate target regions and a single execution of the method is sufficient for different types of targets having different shapes and sizes. Computational efficiency is further increased by parallelizing the algorithm using OpenMP and NVidia CUDA implementations.

A robust nonlinear scale space change detection approach for SAR images

In this paper, we propose a change detection approach based on nonlinear scale space analysis of change images for robust detection of various changes incurred by natural phenomena and/or human activities in Synthetic Aperture Radar (SAR) images using Maximally Stable Extremal Regions (MSERs). To achieve this, a variant of the log-ratio image of multitemporal images is calculated which is followed by Feature Preserving Despeckling (FPD) to generate nonlinear scale space images exhibiting different trade-offs in terms of speckle reduction and shape detail preservation. MSERs of each scale space image are found and then combined through a decision level fusion strategy, namely “selective scale fusion” (SSF), where contrast and boundary curvature of each MSER are considered. The performance of the proposed method is evaluated using real multitemporal high resolution TerraSAR-X images and synthetically generated multitemporal images composed of shapes with several orientations, sizes, and backscatter amplitude levels representing a variety of possible signatures of change. One of the main outcomes of this approach is that different objects having different sizes and levels of contrast with their surroundings appear as stable regions at different scale space images thus the fusion of results from scale space images yields a good overall performance.

Hyperspectral images and LiDAR based DEM fusion: A multi-modal landuse classification strategy

Hyperspectral imaging based land cover/land use classification accuracy is expected to be improved by fusion with a LIDAR based Digital Elevation Model (DEM). To this end, we propose a multi-modal architecture, as well as a filtering technique extracting a shadow invariant one dimensional feature from a pixel spectrum. The proposed approach allows treating shadow and non-shadow areas separately. DEM is incorporated into this architecture through feature extraction and post classification procedures. A digital terrain model estimated from DEM is used to calculate object heights. Slope, curvature and polynomial surface fitting based features are extracted in different scales. In post classification, DEM segments and relatively high objects obtained from DEM are interpreted by superposition with the class map.

Noise estimation for hyperspectral imagery using spectral unmixing and synthesis

Most hyperspectral image (HSI) processing algorithms assume a signal to noise ratio model in their formulation which makes them dependent on accurate noise estimation. Many techniques have been proposed to estimate the noise. A very comprehensive comparative study on the subject is done by Gao et al. [1]. In a nut-shell, most techniques are based on the idea of calculating standard deviation from assumed-to-be homogenous regions in the image. Some of these algorithms work on a regular grid parameterized with a window size w, while others make use of image segmentation in order to obtain homogenous regions. This study focuses not only to the statistics of the noise but to the estimation of the noise itself. A noise estimation technique motivated from a recent HSI de-noising approach [2] is proposed in this study. The denoising algorithm is based on estimation of the end-members and their fractional abundances using non-negative least squares method. The end-members are extracted using the well-known simplex volume optimization technique called NFINDR after manual selection of number of end-members and the image is reconstructed using the estimated endmembers and abundances. Actually, image de-noising and noise estimation are two sides of the same coin: Once we denoise an image, we can estimate the noise by calculating the difference of the de-noised image and the original noisy image. In this study, the noise is estimated as described above. To assess the accuracy of this method, the methodology in [1] is followed, i.e., synthetic images are created by mixing end-member spectra and noise. Since best performing method for noise estimation was spectral and spatial de-correlation (SSDC) originally proposed in [3], the proposed method is compared to SSDC. The results of the experiments conducted with synthetic HSIs suggest that the proposed noise estimation strategy outperforms the existing techniques in terms of mean and standard deviation of absolute error of the estimated noise. Finally, it is shown that the proposed technique demonstrated a robust behavior to the change of its single parameter, namely the number of end-members.

A region based target detection method for SAR images

Automatic target detection methods for synthetic aperture radar (SAR) images are sensitive to image resolution, size of the target to be detected, clutter complexity, and speckle noise level. A robust automatic target detection method needs to be less sensitive to the above factors. In this study, a constant false alarm rate (CFAR) based automatic target detection method which can find a target and its heterogeneous clutter independent of the image resolution and the target size has been developed. The proposed method provides efficient memory usage and low computational complexity.

Interactive ship segmentation in SAR images

Ship detection from synthetic aperture radar (SAR) images is important for various automatic target recognition (ATR) tasks. Although the ships in offshore areas can be easily detected, the ones near the shores or close to each other are difficult to detect. Furthermore, segmentation and classification of such ships is extremely difficult. In this study, a novel approach is presented for the fast and accurate segmentation of ship boundaries with minimal user interaction. In this approach, the rough location and orientation of a ship is determined by the user. Then, a ship model, which is constructed from synthetic ship images, is fitted on to the ship selected by the user and accurate ship boundaries are extracted. The effectiveness of the proposed algorithm is demonstrated by experimental results.

Destriping of hyperion images

Pushbroom hyperspectral images (HSIs) suffer from many unwanted effects such as stripes, smile, random noise etc. Among these phenomena striping is often the first one to be processed in most existing HSI processing chains. An overview of well-known systems as well as recent algorithms for destripping is provided in this paper. A novel destripping technique is proposed. The method is based on the idea of equalizing detector responses. To this end, homogenous lines are detected. Visual examination of the results as well as objective criteria suggest that the proposed technique remove stripes and restore spectral information.

Estimation of noise model parameters for images taken by a full-frame hyperspectral camera

Noise has to be taken into account in the algorithms of classification, target detection and anomaly detection. Recent studies indicate that noise estimation is also crucial in subspace identification of HSI. Several techniques were proposed for noise estimation including: multiple linear regression based techniques, spectral unmixing and remixing etc. The noise in HSI is widely accepted to be a spatially stationary random process. But the variance of the noise varies from one wavelength to another. Two types of noise are considered: the first one is the circuitry noise (thermal noise) which is signal independent. The second one is the photonic noise (shot noise) which is signal dependent. The latter is considered to be the dominant one. A reliable way to accurately estimate the noise requires the identification of a large uniform region in the image. To this end, we propose a region growing technique. At the end of this process, a certain number of regions with different sizes and uniformities are obtained. The next step consists of identifying the most uniform region having the largest area. Once the most uniform and largest region of the scene is identified the next step is to apply an ideal low pass filter to this region. This yields an estimate of the noise-free data, hence the noise itself by calculating the difference. It is also possible to apply the well-known scatter plot technique. Experiments suggest that the proposed scheme produces comparable results to its competitors. A major advantage of the technique is the automated identification of an homogenous region.

Information theoretic SAR boundary detection with user interaction

Detection of region boundaries is a very challenging task especially in the presence of noise or speckle as in synthetic aperture radar images. In this work, we propose a user interaction based boundary detection technique which makes use of B-splines and well-known powerful tools of information theory such as the Kullback-Leibler divergence (KLD) and Bhattacharyya distance. The proposed architecture consists of the following four main steps: (1) The user selects points inside and outside of a region. (2) Profiles that link these inside and outside points are extracted. (3) Boundary points that lie on the profile are located. (4) Finally, the B-splines that provide both elasticity and smoothness are used connect boundary points together to obtain an accurate estimate of the actual boundary. Existing work related to this approach are extended in several axes. First the use of multiple points both inside and outside of a region made possible to obtain a few times more boundary points. A tracking stage is proposed to put the boundary points in the right order and at the same time eliminate some of them that are erroneously detected as boundary points as well. Experiments were conducted using simulated and real SAR images.