Bhabesh Deka

Removal of correlated speckle noise using sparse and overcomplete representations

Abstract Recently, there has been a growing interest in the sparse representation of signals over learned and overcomplete dictionaries. Instead of using fixed transforms such as the wavelets and its variants, an alternative way is to train a redundant dictionary from the image itself. This paper presents a novel de-speckling scheme for medical ultrasound and speckle corrupted photographic images using the sparse representations over a learned overcomplete dictionary. It is shown that the proposed algorithm can be used effectively for the removal of speckle by combining an existing pre-processing stage before an adaptive dictionary could be learned for sparse representation. Extensive simulations are carried out to show the effectiveness of the proposed filter for the removal of speckle noise both visually and quantitatively.

A linear prediction based switching median filter for the removal of salt and pepper noise from highly corrupted image

In this paper, we propose a switching based median filter for preserving the image details while reducing the streaking problem in gray scale images corrupted by salt and pepper noise at high noise ratios. It effectively suppresses the noise in two stages. First, the noisy pixels are detected by using the signal dependent rank-ordered mean (SD-ROM) filter. In the second stage, the noisy pixels are first substituted by the first-order 1D causal linear prediction technique and subsequently replaced by the median value. Extensive simulations are carried out to validate our claim. Experimental results show improvements both visually and quantitatively compared to that of the state-of-the art switching based median filters at high noise ratios.

Single image super-resolution using compressive sensing with learned overcomplete dictionary

This paper proposes a novel framework that unifies the concept of sparsity of a signal over a properly chosen basis set and the theory of signal reconstruction via compressed sensing in order to obtain a high-resolution image derived by using a single down-sampled version of the same image. First, we enforce sparse overcomplete representations on the low-resolution patches of the input image. Then, using the sparse coefficients as obtained above, we reconstruct a high-resolution output image. A blurring matrix is introduced in order to enhance the incoherency between the sparsifying dictionary and the sensing matrices which also resulted in better preservation of image edges and other textures. When compared with the similar techniques, the proposed method yields much better result both visually and quantitatively.

Sparse regularization method for the detection and removal of random-valued impulse noise

In this paper, we propose a novel two-stage algorithm for the detection and removal of random-valued impulse noise using sparse representations. The main aim of the paper is to demonstrate the strength of image inpainting technique for the reconstruction of images corrupted by random-valued impulse noise at high noise densities. First, impulse locations are detected by applying the combination of sparse denoising and thresholding, based on sparse and overcomplete representations of the corrupted image. This stage optimally selects threshold values so that the sum of the number of false alarms and missed detections obtained at a particular noise level is the minimum. In the second stage, impulses, detected in the first stage, are considered as the missing pixels or holes and subsequently these holes are filled-up using an image inpainting method. Extensive simulation results on standard gray scale images show that the proposed method successfully removes random-valued impulse noise with better preservation of edges and other details compared to the existing techniques at high noise ratios.

A Practical Under-Sampling Pattern for Compressed Sensing MRI

Typically, magnetic resonance (MR) images are stored in k-space where the higher energy samples, i.e., the samples with maximum information are concentrated near the center only; whereas, relatively lower energy samples are present near the outer periphery. Recently, variable density (VD) random under-sampling patterns have been increasingly popular and a topic of active research in compressed sensing (CS)-based MR image reconstruction. In this paper, we demonstrate a simple approach to design an efficient k-space under-sampling pattern, namely, the VD Poisson Disk (VD-PD) for sampling MR images in k-space and then implementing the same for CS-MRI reconstruction. Results are also compared with those obtained from some of the most prominent and commonly used sampling patterns, including the VD random with estimated PDF (VD-PDF), the VD Gaussian density (VD-Gaus), the VD uniform random (VD-Rnd), and the Radial Type in the CS-MRI literature.

Wavelet-based Despeckling of Medical Ultrasound Images

Abstract This paper presents a new wavelet-domain technique for despeckling of medical ultrasound (US) images for improved clinical diagnosis. The method uses the generalized Gaussian distribution and generalized gamma distribution to model the image and the speckle, respectively, in the detailed sub-bands of wavelet decomposition of the log-transformed US image. Combining these, a priori distributions with the Bayesian maximum a posteriori criterion, shrinkage estimators are derived for processing the wavelet coefficients of the detail sub-bands. The visual comparison of despeckled US images and the higher values of quality metrics indicate that the new method suppresses the speckle noise well while preserving the texture and organ surfaces.

Random-Valued Impulse Denoising Using a Fast l 1 -Minimization-Based Image Inpainting Technique

In this chapter, an image inpainting approach based on l 1-norm regularization is presented for the estimation of pixels corrupted by the random-valued impulse noise. It is a two-stage reconstruction scheme. First, a reasonably accurate random-valued impulse detection scheme is applied to detect the corrupted pixels. Next, the corrupted pixels are treated as missing pixels and replaced by using an image inpainting technique. The inpainting method is based on the fast iterative shrinkage thresholding algorithm (FISTA). The proposed method is fast and experimental results show that it is robust to non-Gaussian and nonadditive degradations like the random-valued impulse noise. It also outperforms similar random-valued impulse denoising schemes in terms of computational complexity while preserving the image quality.

Weighted Wavelet Tree Sparsity Regularization for Compressed Sensing Magnetic Resonance Image Reconstruction

Compressed sensing in magnetic resonance imaging (CS-MRI) improves the MRI scan time by acquiring only a few k-space samples and then reconstructs the image using a nonlinear procedure from the highly undersampled measurements. Besides the standard wavelet sparsity, MR images are also found to exhibit tree sparsity across various scales of the wavelet decomposition which are generally modeled as overlapping group sparsity. In this chapter, we propose a novel iteratively weighted wavelet tree sparsity based CS-MRI reconstruction technique to estimate MR images from highly undersampled Fourier measurements. Simulations on various real MR images show that the proposed technique offers significant improvements compared to the state-of-the-art either in terms of visual quality or k-space measurements with the same reconstruction time.

Efficient interpolated compressed sensing reconstruction scheme for 3D MRI

3D magnetic resonance imaging (3D MRI) is one of the most preferred medical imaging modalities for the analysis of anatomical structures where acquisition of multiple slices along the slice select gradient direction is very common. In 2D multi-slice acquisition, adjacent slices are highly correlated because of very narrow inter-slice gaps. Application of compressed sensing (CS) in MRI significantly reduces traditional MRI scan time due to random undersampling. The authors first propose a fast interpolation technique to estimate missing samples in the k-space of a highly undersampled slice (H-slice) from k-space (s) of neighbouring lightly undersampled slice/s (L-slice). Subsequently, an efficient multislice CS-MRI reconstruction technique based on weighted wavelet forest sparsity, and joint total variation regularisation norms is applied simultaneously on both interpolated H and non-interpolated L-slices. Simulation results show that the proposed CS reconstruction for 3D MRI is not only computationally faster but significant improvements in terms of visual quality and quantitative performance metrics are also achieved compared to the existing methods.

A Fast Satellite Image Super-Resolution Technique Using Multicore Processing

Sparse representation based single image super-resolution technique requires several regularization problems to be solved to generate the desired output. It is computationally intensive and needs a considerable time if we implement sequentially on a single core processor. Remote sensing applications generally require high resolution satellite images on a near real-time instant. Since, satellite images are of larger dimensions, so obtaining desired high resolution images within some practical time will be highly data intensive. Therefore, fast super-resolution based post-processing may be integrated into the existing system either in software or hardware for practical applications. In this paper, we implement an OpenMP based parallel processing technique for single image super-resolution of multispectral satellite images. Results not only show a promising speed up in the execution time but provide visually enhanced outputs as well, compared to some of the existing methods.