Mehran Yazdi

Compression of Hyperspectral Images Using Discerete Wavelet Transform and Tucker Decomposition

The compression of hyperspectral images (HSIs) has recently become a very attractive issue for remote sensing applications because of their volumetric data. In this paper, an efficient method for hyperspectral image compression is presented. The proposed algorithm, based on Discrete Wavelet Transform and Tucker Decomposition (DWT-TD), exploits both the spectral and the spatial information in the images. The core idea behind our proposed technique is to apply TD on the DWT coefficients of spectral bands of HSIs. We use DWT to effectively separate HSIs into different sub-images and TD to efficiently compact the energy of sub-images. We evaluate the effect of the proposed method on real HSIs and also compare the results with the well-known compression methods. The obtained results show a better performance of the proposed method. Moreover, we show the impact of compression HSIs on the supervised classification and linear unmixing.

Panchromatic and Multispectral Image Fusion Based on Maximization of Both Spectral and Spatial Similarities

The panchromatic (PAN) sharpening of multispectral (MS) bands can be performed by fusing the PAN and MS images. Measuring similarity criterion computed among input images is one way to synthesize MS images in higher resolution based on either spectral or spatial domains. However, a few methods consider both spectral and spatial similarities. In this paper, the fusion between PAN and MS images is performed by engaging both similarities. We use the spectral histogram, recently introduced to characterize the spectral information of an image in different frequency ranges, as the spectral similarity criterion. This similarity suggests considering a statistical similarity measure between two spectral histograms of two images. Furthermore, we use the fourth-order correlation coefficient as a spatial similarity criterion instead of correlation coefficient. Meanwhile, in the decision level of fusion process, a proper threshold should be selected to determine whether the details should be injected or not. There is no reference to choose it in general cases, and this threshold is calculated for each set of input images separately and is based on intersecting two similarity curves. We do this by first calculating the spatial and spectral similarity criteria for some specific threshold values and then fit two similarity curves on these sample points by the spline interpolation method. Then, after decomposing input images using the nonsubsampled contourlet transform, we inject the PAN details into the MS details considering the selected threshold. The experimental results obtained by applying the proposed image fusion method indicate some improvements in the fusion performance.

Noise Reduction of Hyperspectral Images Using Kernel Non-Negative Tucker Decomposition

We propose a new noise reduction algorithm for the denoising of hyperspectral images. The proposed algorithm, Genetic Kernel Tucker Decomposition (GKTD), exploits both the spectral and the spatial information in the images. With respect to a previous approach, we use the kernel trick to apply a Tucker decomposition on a higher dimensional feature space instead of the input space. A genetic algorithm is used to optimize for the lower rank Tucker tensor in the feature space. We evaluate the effect of the kernel algorithm with respect to non-kernel GTD, and also compare the results to those from principal component analysis bivarate wavelet shirinking on real images. Our results show a better performance of the proposed method.

A Novel Image Fusion Method Using Curvelet Transform Based on Linear Dependency Test

Because of the benefits of image fusion, although higher resolution remote sensing data are available now, image fusion is still a popular method for better interpreting image data. This paper focuses on a novel region-based image fusion method which facilitates increased flexibility with the definition of a variety of fusion rules. To do that, we use the curvelet transform to merge the details of images. Also, we introduce a fusion rule decision based on the linear algebra that helps to do a better fusion of detail coefficients of the curvelet transform. The experimental results show improvement of the proposed method compared with the well-known methods.

Steel Surface Defect Detection Using Texture Segmentation Based on Multifractal Dimension

Recently, it becomes significant to enhance quality of products as well as to increase quantity of products in the steel manufacturing industry. As a manufacturing gets faster, the fast and exact detection of defect is important to acquire a competitive power. Without automatic machine vision technology, steel rolling operations is not able to perform real-time inline surface defect inspection. In this paper, we propose a new defect detection algorithm based on multifractal. Then, some suitable features are extracted and presented to neural network for classification. The obtained accuracy is % 97.9.

An efficient algorithm for function optimization: modified stem cells algorithm

In this paper, we propose an optimization algorithm based on the intelligent behavior of stem cell swarms in reproduction and self-organization. Optimization algorithms, such as the Genetic Algorithm (GA), Particle Swarm Optimization (PSO) algorithm, Ant Colony Optimization (ACO) algorithm and Artificial Bee Colony (ABC) algorithm, can give solutions to linear and non-linear problems near to the optimum for many applications; however, in some case, they can suffer from becoming trapped in local optima. The Stem Cells Algorithm (SCA) is an optimization algorithm inspired by the natural behavior of stem cells in evolving themselves into new and improved cells. The SCA avoids the local optima problem successfully. In this paper, we have made small changes in the implementation of this algorithm to obtain improved performance over previous versions. Using a series of benchmark functions, we assess the performance of the proposed algorithm and compare it with that of the other aforementioned optimization algorithms. The obtained results prove the superiority of the Modified Stem Cells Algorithm (MSCA).

An opposite view data replacement approach for reducing artifacts due to metallic dental objects.

PURPOSE To present a conceptually new method for metal artifact reduction (MAR) that can be used on patients with multiple objects within the scan plane that are also of small sized along the longitudinal (scanning) direction, such as dental fillings. METHODS The proposed algorithm, named opposite view replacement, achieves MAR by first detecting the projection data affected by metal objects and then replacing the affected projections by the corresponding opposite view projections, which are not affected by metal objects. The authors also applied a fading process to avoid producing any discontinuities in the boundary of the affected projection areas in the sinogram. A skull phantom with and without a variety of dental metal inserts was made to extract the performance metric of the algorithm. A head and neck case, typical of IMRT planning, was also tested. RESULTS The reconstructed CT images based on this new replacement scheme show a significant improvement in image quality for patients with metallic dental objects compared to the MAR algorithms based on the interpolation scheme. For the phantom, the authors showed that the artifact reduction algorithm can efficiently recover the CT numbers in the area next to the metallic objects. CONCLUSIONS The authors presented a new and efficient method for artifact reduction due to multiple small metallic objects. The obtained results from phantoms and clinical cases fully validate the proposed approach.

A novel approach for reducing metal artifacts due to metallic dental implants

We present a conceptually new method for metal artifact reduction (MAR). This method can be used when a patient has multiple metal objects with small sizes. The proposed algorithm uses raw CT data (sinogram) instead of reconstructed CT slices. First, the projection data affected by metal objects (missing projections) are detected in sinogram. Then, the missing projections are replaced by corresponding unaffected projections in other slices or opposite view. Finally, the modified sinogram is transferred back to the CT scanner device where CT slices are regenerated using the built-in reconstruction operator. The obtained results on real patients demonstrate the superiority of our method compared with the traditional projection-interpolation method.

A robust clustering method based on blind, naked mole-rats (BNMR) algorithm

Abstract One of most popular data clustering algorithms is K-means algorithm that uses the distance criterion for measuring the correlation among data. To do that, we should know in advance the number of classes (K) and choose K data point as an initial set to run the algorithm. However, the choice of initial points is a main problem in this algorithm, which may cause that the algorithm converges to local optima. So, some other clustering algorithms have been proposed to overcome this problem such as the methods based on K-means (SBKM), Genetic Algorithm (GAPS and VGAPS), Particle Swarm Optimization (PSO), Ant Colony Optimization (Dynamic ants), Simulated Annealing (SA) and Artificial Bee Colony (ABC) algorithm. In this paper, we employ a new meta-heuristic algorithm. We called it blind, naked mole-rats (BNMR) algorithm, for data clustering. The algorithm was inspired by social behavior of the blind, naked mole-rats colony in searching the food and protecting the colony against invasions. We developed a new data clustering based on this algorithm, which has the advantages such as high speed of convergence. The experimental results obtained by using the new algorithm on different well-known datasets compared with those obtained using other mentioned methods showed the better accuracy and high speed of the new algorithm.

A powerful and efficient evolutionary optimization algorithm based on stem cells algorithm for data clustering

There are many ways to divide datasets into some clusters. One of most popular data clustering algorithms is K-means algorithm which uses the distance criteria for measuring the data correlation. To do that, we should know in advance the number of classes (K) and choose K data points as an initial set to run the algorithm. However, the choice of initial points is a main problem in this algorithm which may cause the algorithm to converge to a local minimum. Some other data clustering algorithms have been proposed to overcome this problem. The methods are Genetic algorithm (GA), Ant Colony Optimization (ACO), PSO algorithm, and ABC algorithms. In this paper, we employ the Stem Cells Optimization algorithm for data clustering. The algorithm was inspired by behavior of natural stem cells in the human body. We developed a new data clustering based on this new optimization scheme which has the advantages such as high convergence rate and easy implementation process. It also avoids local minimums in an intelligent manner. The experimental results obtained by using the new algorithm on different well-known test datasets compared with those obtained using other mentioned methods demonstrate the better accuracy and high speed of the new algorithm.