High-resolution Dictionary Research Articles

Dual-dictionary learning based MR image reconstruction with self-adaptive dictionaries.

Dual-dictionary learning method utilizes two dictionaries at two different resolution levels, a high resolution dictionary trained with full-data training set, and a low resolution dictionary co-trained with corresponding undersampled dataset. This method effectively incorporates a priori knowledge of typical structures, specific features and local details, leading to its success in magnetic resonance (MR) image reconstruction from highly undersampled k-space data. In this paper, we improve this dual-dictionary learning method by using self-adaptive dictionaries. The two level dictionaries are updated correspondingly in the inner iteration after updating the reconstruction result to maintain their matching accuracy. Experimental results show that the proposed method can improve the reconstruction quality efficiently and enhance the robustness significantly.

Single‐image super‐resolution in RGB space via group sparse representation

Super-resolution (SR) is the problem of generating a high-resolution (HR) image from one or more low-resolution (LR) images. This study presents a new approach to single-image super-resolution based on group sparse representation. Two dictionaries are constructed corresponding to the LR and HR image patches, respectively. The sparse coefficients of an input LR image patch in terms of the LR dictionary are used to recover the HR patch from the HR dictionary. When constructing the dictionaries, the three colour channels in a training image patch are considered a group composed of three atoms. The whole group is selected simultaneously when representing an image patch so that the correlations between the colour channels can be retained. A dictionary training method is also designed in which the two dictionaries are trained jointly to ensure that the corresponding LR and HR patches have the same sparse coefficients. Experimental results demonstrate the effectiveness of the proposed method and its robustness to noise.

Image super-resolution via sparse representation over multiple learned dictionaries based on edge sharpness

A new algorithm for single-image super-resolution based on selective sparse representation over a set of coupled dictionary pairs is proposed. Patch sharpness measure for high- and low-resolution patch pairs defined via the magnitude of the gradient operator is shown to be approximately invariant to the patch resolution. This measure is employed in the training stage for clustering the training patch pairs and in the reconstruction stage for model selection. For each cluster, a pair of low- and high-resolution dictionaries is learned. In the reconstruction stage, the sharpness measure of a low-resolution patch is used to select the cluster it belongs to. The sparse coding coefficients of the patch over the selected low-resolution cluster dictionary are calculated. The underlying high-resolution patch is reconstructed by multiplying the high-resolution cluster dictionary with the calculated coefficients. The performance of the proposed algorithm is tested over a set of natural images. PSNR and SSIM results show that the proposed algorithm is competitive with the state-of-the-art super-resolution algorithms. In particular, it significantly out-performs the state-of-the-art algorithms for images with sharp edges and corners. Visual comparison results also support the quantitative results.

Accelerating MRI reconstruction via three-dimensional dual-dictionary learning using CUDA

Using undersampled k-space data for reconstruction is an effective way to accelerate data acquisition of magnetic resonance imaging (MRI). With the development of compressed sensing (CS) theory, many solutions have been proposed for undersampled data reconstruction. Moreover, dictionary learning method has shown good results in improving reconstruction quality. However, CS reconstruction algorithms are time consuming, especially at dictionary training and sparse coding step. The computation overhead is even higher for three-dimensional reconstruction. With a large number of slices, data size can be massive and more time consuming. In this paper, we use three-dimensional dual-dictionary learning scheme for the reconstruction procedure. Three-dimensional dictionaries train the dictionary atoms in image blocks and utilize spatial correlation among MR slices. Dual-dictionary learning method cooperates low-resolution dictionary and high-resolution dictionary for sparse coding and image updating, respectively. Compute unified device architecture (CUDA) is utilized to design the parallel algorithms on graphics processing unit (GPU). We mainly optimize dictionary learning algorithm and image updating. We also develop parallel CPU codes using OpenMP (Open Multi-Processing) and another version of CUDA codes with algorithmic optimization. Experimental results show that more than 324 times of speedup is achieved compared with CPU-only codes with 24 MRI slices and more than 40 times of acceleration compared with OpenMP codes.

Hyperspectral imagery super-resolution by compressive sensing inspired dictionary learning and spatial-spectral regularization.

Due to the instrumental and imaging optics limitations, it is difficult to acquire high spatial resolution hyperspectral imagery (HSI). Super-resolution (SR) imagery aims at inferring high quality images of a given scene from degraded versions of the same scene. This paper proposes a novel hyperspectral imagery super-resolution (HSI-SR) method via dictionary learning and spatial-spectral regularization. The main contributions of this paper are twofold. First, inspired by the compressive sensing (CS) framework, for learning the high resolution dictionary, we encourage stronger sparsity on image patches and promote smaller coherence between the learned dictionary and sensing matrix. Thus, a sparsity and incoherence restricted dictionary learning method is proposed to achieve higher efficiency sparse representation. Second, a variational regularization model combing a spatial sparsity regularization term and a new local spectral similarity preserving term is proposed to integrate the spectral and spatial-contextual information of the HSI. Experimental results show that the proposed method can effectively recover spatial information and better preserve spectral information. The high spatial resolution HSI reconstructed by the proposed method outperforms reconstructed results by other well-known methods in terms of both objective measurements and visual evaluation.

Accelerating the reconstruction of magnetic resonance imaging by three-dimensional dual-dictionary learning using CUDA.

An effective way to improve the data acquisition speed of magnetic resonance imaging (MRI) is using under-sampled k-space data, and dictionary learning method can be used to maintain the reconstruction quality. Three-dimensional dictionary trains the atoms in dictionary in the form of blocks, which can utilize the spatial correlation among slices. Dual-dictionary learning method includes a low-resolution dictionary and a high-resolution dictionary, for sparse coding and image updating respectively. However, the amount of data is huge for three-dimensional reconstruction, especially when the number of slices is large. Thus, the procedure is time-consuming. In this paper, we first utilize the NVIDIA Corporation's compute unified device architecture (CUDA) programming model to design the parallel algorithms on graphics processing unit (GPU) to accelerate the reconstruction procedure. The main optimizations operate in the dictionary learning algorithm and the image updating part, such as the orthogonal matching pursuit (OMP) algorithm and the k-singular value decomposition (K-SVD) algorithm. Then we develop another version of CUDA code with algorithmic optimization. Experimental results show that more than 324 times of speedup is achieved compared with the CPU-only codes when the number of MRI slices is 24.

Synthetic image super resolution using FeatureMatch

In this paper, we propose a super resolution (SR) method for synthetic images using FeatureMatch. Existing state-of-the-art super resolution methods are learning based methods, where a pair of low-resolution and high-resolution dictionary pair are trained, and this trained pair is used to replace patches in low-resolution image with appropriate matching patches from the high-resolution dictionary. In this paper, we show that by using Approximate Nearest Neighbour Fields (ANNF), and a common source image, we can by-pass the learning phase, and use a single image for dictionary. Thus, reducing the dictionary from a collection obtained from hundreds of training images, to a single image. We show that by modifying the latest developments in ANNF computation, to suit super resolution, we can perform much faster and more accurate SR than existing techniques. To establish this claim we will compare our algorithm against various state-of-the-art algorithms, and show that we are able to achieve better and faster reconstruction without any training phase.

Image Super-Resolution via Low-Rank Representation

This paper presents a novel method for solving single-image super-resolution problems, based upon low-rank representation (LRR). Given a set of a low-resolution image patches, LRR seeks the lowest-rank representation among all the candidates that represent all patches as the linear combination of the patches in a low-resolution dictionary. By jointly training two dictionaries for the low-resolution and high-resolution images, we can enforce the similarity of LLRs between the low-resolution and high-resolution image pair with respect to their own dictionaries. Therefore, the LRR of a low-resolution image can be applied with the high-resolution dictionary to generate a high-resolution image. Unlike the well-known sparse representation, which computes the sparsest representation of each image patch individually, LRR aims at finding the lowest-rank representation of a collection of patches jointly. LRR better captures the global structure of image. Experiments show that our method gives good results both visually and quantitatively.

Structural similarity based single image super-resolution with nonlocal regularization

Recently, sparse coding based image super-resolution has attracted increasing interests. This paper proposes an improved image super-resolution method, by incorporating structural similarity (SSIM) index and nonlocal regularization into the framework of image super-resolution via sparse coding. Firstly, an algorithm of combining SSIM based sparse coding and K-SVD is proposed to train the high resolution (HR) and low resolution (LR) dictionary pairs. And then, the sparse representations of observed LR image are sought to reconstruct the HR image with the trained LR and HR dictionary pairs by exploiting nonlocal self-similarities. Experimental results demonstrate the effectiveness of the proposed method, both in its visual effects and in quantitative terms.

Infrared image super-resolution via locality-constrained group sparse model

Aiming at the problems of low-resolution and poor visual quality of infrared images, a locality-constrained group sparsity based infrared image super-resolution algorithm is proposed. Firstly with considering the texture self-similarity of infrared images and group structural sparsity of atom coefficients, a locality-constrained group sparse (LCGS) model is proposed. Secondly, under LCGS and K-singular value decomposition, a pair of group structural dictionaries is learned. The dictionary pair can well capture and preserve the intrinsic geometrical manifold of low and high resolution data. Finally, the high-resolution infrared images are recovered by the high-resolution dictionary and the corresponding low-resolution group sparse coefficients. Experimental results show that the proposed method obtains excellent performance in objective evaluation and subjective visual effect.

Denoising and fast diffusion imaging with physically constrained sparse dictionary learning

Diffusion-weighted imaging (DWI) allows imaging the geometry of water diffusion in biological tissues. However, DW images are noisy at high b-values and acquisitions are slow when using a large number of measurements, such as in Diffusion Spectrum Imaging (DSI). This work aims to denoise DWI and reduce the number of required measurements, while maintaining data quality. To capture the structure of DWI data, we use sparse dictionary learning constrained by the physical properties of the signal: symmetry and positivity. The method learns a dictionary of diffusion profiles on all the DW images at the same time and then scales to full brain data. Its performance is investigated with simulations and two real DSI datasets. We obtain better signal estimates from noisy measurements than by applying mirror symmetry through the q-space origin, Gaussian denoising or state-of-the-art non-local means denoising. Using a high-resolution dictionary learnt on another subject, we show that we can reduce the number of images acquired while still generating high resolution DSI data. Using dictionary learning, one can denoise DW images effectively and perform faster acquisitions. Higher b-value acquisitions and DSI techniques are possible with approximately 40 measurements. This opens important perspectives for the connectomics community using DSI.

Reconstruction of magnetic resonance imaging by three‐dimensional dual‐dictionary learning

To improve the magnetic resonance imaging (MRI) data acquisition speed while maintaining the reconstruction quality, a novel method is proposed for multislice MRI reconstruction from undersampled k-space data based on compressed-sensing theory using dictionary learning. There are two aspects to improve the reconstruction quality. One is that spatial correlation among slices is used by extending the atoms in dictionary learning from patches to blocks. The other is that the dictionary-learning scheme is used at two resolution levels; i.e., a low-resolution dictionary is used for sparse coding and a high-resolution dictionary is used for image updating. Numerical experiments are carried out on in vivo 3D MR images of brains and abdomens with a variety of undersampling schemes and ratios. The proposed method (dual-DLMRI) achieves better reconstruction quality than conventional reconstruction methods, with the peak signal-to-noise ratio being 7 dB higher. The advantages of the dual dictionaries are obvious compared with the single dictionary. Parameter variations ranging from 50% to 200% only bias the image quality within 15% in terms of the peak signal-to-noise ratio. Dual-DLMRI effectively uses the a priori information in the dual-dictionary scheme and provides dramatically improved reconstruction quality.

An improved joint dictionary training method for single image super resolution

PurposeThe purpose of this paper is to introduce an improved method for joint training of low‐ and high‐resolution dictionaries in the single image super resolution. With simulations, the proposed method is thereafter evaluated.Design/methodology/approachSparse representations of low‐resolution image patches are used to reconstruct the high‐resolution image patches with high resolution dictionary. By using different factors, the scheme weights the two dictionaries in the high‐ and low‐resolution spaces in the training process. It is reasonable to achieve better reconstructed images with more emphasis on the high‐resolution spaces.FindingsAn improved joint training algorithm based on K‐SVD is developed with flexible weight factors on dictionaries in the high‐ and low‐resolution spaces. From the experiment results, the proposed scheme outperforms the classic bicubic interpolation and neighbor‐embedding learning based method.Originality/valueBy using flexible weight factors in joint training of the dictionaries for super resolution, better reconstruction results can be achieved.

Enhancing dictionary-based super-resolution using nonlocal total variation regularization

The dictionary-based super-resolution (SR) method has received much attention in recent years because sparse representation is very effective for image restoration tasks. By sparse representation, original image patches are represented as a sparse linear combination of atoms in an over-complete dictionary. However, the dictionary-based SR approach has some disadvantages in that it produces annoying ringing artifacts, especially along the object boundaries and is not effective in reconstructing images that contain patterns with strong edges. We enhance the dictionary-based SR using nonlocal total variation regularization. In the training stage, we jointly train two dictionaries, Dh and D1, from the low-resolution (LR) and high-resolution (HR) training image patches by using the K-singular value decomposition (KSVD) algorithm as in conventional methods. In the synthesis stage, we obtain the sparse coefficient vector from the LR test image over the LR dictionary, and reconstruct SR patches using the coefficient vector over the HR dictionary. Then, we employ nonlocal total variation regularization to remove annoying ringing artifacts and recover the patterns with strong edges in images. Experimental results on various test images demonstrate that the proposed method is very effective in enhancing the dictionary-based SR approaches in terms of quantitative performance and visual quality.

Face Image Superresolution via Locality Preserving Projection and Sparse Coding

It is important to enhance the resolution of face images from video surveillance for recognization and other post processing. In this paper, a novel sparse representation based face image superresolution (SR) method is proposed to reconstruct a high resolution (HR) face image from a LR observation. First, it gets a HR-LR dictionary pair for certain input LR patch via position patch clustering and locality preserving projection (LPP). LPP is used to further refine the training samples by taking into account the geometry structure between training patches. It can make the dictionary more expressive. Second, it sparse codes the input patch over LR dictionary by solving L1-regularized least squares optimization. F eature-sign algorithm is used to get the stable solution. Third, it maps the sparse coding coefficitons obtain from the LR dictionary to the HR dictionary, and reconstruct the HR face image patch with the coefficitions and HR dictionary. By integrating the hallucinated HR patches together with overlapped in adjacent, we can obtain the final HR face image. Experiments conducted on CAS-PEAL-R1 database validate the proposed method both in subjective and objective quality.

Unification of image fusion and super-resolution using jointly trained dictionaries and local information contents

Multi-modal imaging requires image fusion to combine advantages of different types of sensors and requires super-resolution (SR) because of limited spatial resolution of source images. In this study, a novel framework is proposed for unification of image SR and the fusion process to obtain a high-resolution (HR)-fused image from a set of low-resolution (LR) multi-modal images. The jointly trained dictionaries of LR patches and corresponding HR patches are used for sparse representation of LR source image patches and HR-fused image patches, respectively. The sparse coefficients vectors for corresponding patches of source LR images are determined by using orthogonal matching pursuit and a local information content-based metric is employed to fuse these sparse coefficients. The corresponding HR-fused image patch is obtained by combining elements of the HR dictionary as per the fused coefficients of the LR image patches. The experimental results on sets of multi-modal images exhibited that the proposed method outperformed the existing fusion and SR techniques in terms of visual quality and image fusion quality metrics.

Infrared surveillance image super resolution via group sparse representation

Infrared (IR) surveillance image generally has low resolution and signal-to-noise ratios. In this paper, we present a new approach to IR image super resolution via group sparse representation (GS). First based on IR image feature, an algorithm of combining the group orthogonal matching pursuit and K-SVD is proposed to train the dictionaries. The dictionary training can ensure that the corresponding low resolution (LR) and high resolution (HR) image have same GS coefficients. Then the group sparse coefficients of the input LR image are sought to reconstruct the HR image with the trained LR and HR dictionary pair. Experimental results indicate that our method generates sharper results with higher Peak Signal to Noise Ratio (PSNR), and is more robust to noise than several popular methods.

Learning Based Compressed Sensing for SAR Image Super-Resolution

This paper presents a novel approach for the reconstruction of super-resolution (SR) synthetic aperture radar (SAR) images in the compressed sensing (CS) theory framework. Recent research has shown that super-resolved data can be reconstructed from an extremely small set of measurements compared to that currently required. Therefore, a CS to produce SAR super-resolution images is introduced in the present work. The proposed approach contributes in three ways. First, enhanced SR results are achieved using a framework that combines CS with a multi-dictionary. Then, the multi-dictionary pairs are trained after classifying the training images through a sparse coding spatial pyramid machine. Each dictionary pair containing low- and high-resolution dictionaries are jointly trained. Finally, the gradient-descent optimization approach is applied to decrease the mutual coherence between the measurement matrix and the representation basis. The CS reconstruction effect is related to incoherence. The effectiveness of this method is demonstrated on TerraSAR-X data.

SUPER-RESOLUTION OF HYPERSPECTRAL IMAGES USING COMPRESSIVE SENSING BASED APPROACH

Abstract. Over the past decade hyper spectral (HS) image analysis has turned into one of the most powerful and growing technologies in the field of remote sensing. While HS images cover large area at fine spectral resolution, their spatial resolutions are often too coarse for the use in various applications. Hence improving their resolution has a high payoff. This paper presents a novel approach for super-resolution (SR) of HS images using compressive sensing (CS). Besides ill-posedness of SR problem, the main challenge in HS super-resolution is to preserve spectral contents among all bands while increasing their spatial resolutions. In this work, we first obtain an initial estimate of the super-resolution on a reduced dimension HS data. The HS observations of different wavelengths are represented as linear combination of smaller number of basis image planes (BIPs) using principal component analysis (PCA). The novelty of our approach lies in using CS based approach to super-resolve the most informative PCA transformed image representing highest spectral variance (i.e. the first principal component). Our approach uses low and high spatial resolution dictionaries of patches generated by random sampling of raw patches of PCA transformed images that are generated using the training images having similar statistical properties. Using the sparsity constraint, low resolution test patch is represented as a sparse linear combination of relevant dictionary elements adaptively, that leads to initial estimate of super-resolved PCA image having maximum spectral variability. Since SR is an ill-posed problem, we obtain the final solution using a regularization framework considering the sparse coefficients obtained by the CS approach and the autoregressive (AR) parameters obtained from the initial estimate. The remaining PCA images are up-scaled using regularization, considering the same AR parameters which were obtained from super-resolved PCA image having maximum spectral variability. Experiments are conducted on real HS images collected by Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Visual inspections and quantitative comparison confirm that our method enhances spatial information without introducing significant spectral distortion.