K-SVD Dictionary Learning Algorithm Research Articles

Automatic Choroidal Segmentation in Optical Coherence Tomography Images Based on Curvelet Transform and Graph Theory.

Automatic segmentation of the choroid on optical coherence tomography (OCT) images helps ophthalmologists in diagnosing eye pathologies. Compared to manual segmentations, it is faster and is not affected by human errors. The presence of the large speckle noise in the OCT images limits the automatic segmentation and interpretation of them. To solve this problem, a new curvelet transform-based K-SVD method is proposed in this study. Furthermore, the dataset was manually segmented by a retinal ophthalmologist to draw a comparison with the proposed automatic segmentation technique. In this study, curvelet transform-based K-SVD dictionary learning and Lucy-Richardson algorithm were used to remove the speckle noise from OCT images. The Outer/Inner Choroidal Boundaries (O/ICB) were determined utilizing graph theory. The area between ICB and outer choroidal boundary was considered as the choroidal region. The proposed method was evaluated on our dataset and the average dice similarity coefficient (DSC) was calculated to be 92.14% ± 3.30% between automatic and manual segmented regions. Moreover, by applying the latest presented open-source algorithm by Mazzaferri et al. on our dataset, the mean DSC was calculated to be 55.75% ± 14.54%. A significant similarity was observed between automatic and manual segmentations. Automatic segmentation of the choroidal layer could be also utilized in large-scale quantitative studies of the choroid.

Temperature Field Reconstruction Method for Acoustic Tomography Based on Multi-Dictionary Learning.

A reconstruction algorithm is proposed, based on multi-dictionary learning (MDL), to improve the reconstruction quality of acoustic tomography for complex temperature fields. Its aim is to improve the under-determination of the inverse problem by the sparse representation of the sound slowness signal (i.e., reciprocal of sound velocity). In the MDL algorithm, the K-SVD dictionary learning algorithm is used to construct corresponding sparse dictionaries for sound slowness signals of different types of temperature fields; the KNN peak-type classifier is employed for the joint use of multiple dictionaries; the orthogonal matching pursuit (OMP) algorithm is used to obtain the sparse representation of sound slowness signal in the sparse domain; then, the temperature distribution is obtained by using the relationship between sound slowness and temperature. Simulation and actual temperature distribution reconstruction experiments show that the MDL algorithm has smaller reconstruction errors and provides more accurate information about the temperature field, compared with the compressed sensing and improved orthogonal matching pursuit (CS-IMOMP) algorithm, which is an algorithm based on compressed sensing and improved orthogonal matching pursuit (in the CS-IMOMP, DFT dictionary is used), the least square algorithm (LSA) and the simultaneous iterative reconstruction technique (SIRT).

Retinal image assessment using bi-level adaptive morphological component analysis.

The automated analysis of retinal images is a widely researched area which can help to diagnose several diseases like diabetic retinopathy in early stages of the disease. More specifically, separation of vessels and lesions is very critical as features of these structures are directly related to the diagnosis and treatment process of diabetic retinopathy. The complexity of the retinal image contents especially in images with severe diabetic retinopathy makes detection of vascular structure and lesions difficult. In this paper, a novel framework based on morphological component analysis (MCA) is presented which benefits from the adaptive representations obtained via dictionary learning. In the proposed Bi-level Adaptive MCA (BAMCA), MCA is extended to locally deal with sparse representation of the retinal images at patch level whereas the decomposition process occurs globally at the image level. BAMCA method with appropriately offline learnt dictionaries is adopted to work on retinal images with severe diabetic retinopathy in order to simultaneously separate vessels and exudate lesions as diagnostically useful morphological components. To obtain the appropriate dictionaries, K-SVD dictionary learning algorithm is modified to use a gated error which guides the process toward learning the main structures of the retinal images using vessel or lesion maps. Computational efficiency of the proposed framework is also increased significantly through some improvement leading to noticeable reduction in run time. We experimentally show how effective dictionaries can be learnt which help BAMCA to successfully separate exudate and vessel components from retinal images even in severe cases of diabetic retinopathy. In this paper, in addition to visual qualitative assessment, the performance of the proposed method is quantitatively measured in the framework of vessel and exudate segmentation. The reported experimental results on public datasets demonstrate that the obtained components can be used to achieve competitive results with regard to the state-of-the-art vessel and exudate segmentation methods.

Wavelet integrated residual dictionary training for single image super-resolution

Single image super-resolution (SISR) is an encouraging process for image processing applications, where the super-resolution (SR) problem is resolved by the sparse representation of a signal. This paper presents a novel single image SR algorithm based on sparse representation in the wavelet domain. Dictionary learning and sparse coding process for acquiring the high-frequency details into the wavelet domain, wavelet gives better results because wavelets promote the sparsity of the image and also provide structural information about the image. Our method combines the concept of residual image learning with the coupled K-SVD dictionary learning algorithm. Residual image is obtained by subtracting the level-one DWT approximation sub-band from level-two DWT approximation sub-band. The proposed method combined properties of wavelet decomposition such as preserving the high-frequency details, directionality-and compactness by learning the multiple dictionaries for each wavelet subband. Six wavelets integrated residual dictionaries are learned from level-one and level-two wavelet sub-band images, three for the low resolution and three for high-resolution wavelet subband images. By designing compact wavelet multiple dictionaries, we reduce the computational complexity as the number of dictionary atoms is small. The proposed algorithm is tested over four different datasets and compared with eight other super-resolution algorithms in terms of average Peak signal-to-noise ratio (PSNR) and Structural similarity index measure (SSIM). We also examine how super-resolved image can be affected by the dictionary size and patch size as these parameters play a very important role in reconstructing the HR images.

Learning recurrent residual regressors for single image super-resolution

Abstract Example regression-based single image super-resolution (SR) technique has been recognized as an effective way to produce a high-quality image with finer details from one low-resolution (LR) input. However, most current popular approaches usually establish the mappings from the LR feature space to the final HR one in one-pass scheme, which is insufficient to represent the complicated mapping relationship well. In this paper, we propose a novel single image SR framework by learning a group of linear residual regressors in a boosting manner so as to alleviate the gap between the underlying mappings and estimated mappings. In the training stage, we begin with the learning of a set of linear regressors by integrating the K-SVD dictionary learning algorithm and the ridge regression, and then further improve the HR estimate accuracy by learning multi-round residual regressors from the estimated errors in a cascade manner. Accordingly, in the testing stage more details can be gradually added into the input LR image by applying the learned multi-round residual regressors to SR reconstruction. The proposed SR method is fundamentally coarse-to-fine. Experimental results carried out on six publicly available datasets indicate that the proposed SR framework achieves promising performance in comparing with other state-of-the-art competitors in terms of both subjective and objective equality assessments.

Single image super resolution based on sparse representation using discrete wavelet transform

Single image super resolution (SR) based on sparse representation is a promising technique where the SR problem is solved by searching for the most robust representation of a signal in terms of atoms in a dictionary. However, first and second-order derivatives are always used as features for patches to be trained as dictionaries and super-resolved patches are reconstructed using dictionaries and sparse representation of these features. In this paper, a novel single image SR algorithm based on sparse representation with considering the effect of significant features is proposed. Therefore, high frequency details are preserved using discrete wavelet transform (DWT). In addition, an intermediate process is also proposed to learn finer dictionaries and thereby estimate the sharper and more detailed super-resolved image. The dictionaries are constructed from the distinctive features using K-SVD dictionary learning algorithm. The intermediate process uses approximation subband. Thus, constructed dictionaries contain so much more significant information and the interpolated high frequency components are corrected. Therefore, the intermediate process restores the high frequency details better in super-resolved images. The proposed algorithm was tested on ‘Set14’ dataset. Owing to DWT, the proposed algorithm recovers the edges better as well as improving the computational efficiency. The quantitative and visual results show the superiority and competitiveness of the proposed method over the simplest techniques and state-of-art SR algorithms. Experimental time comparisons with the state-of-art algorithms validate the effectiveness of the proposed approach.

Geophysical inversion by dictionary learning

Dictionary learning, a form of unsupervised machine learning, has recently been applied to ocean sound speed profile (SSP) data to obtain compact dictionaries of shape functions which explain SSPs using as few as one non-zero coefficient. In this presentation, the results of this analysis and potential applications of dictionary learning techniques to the inversion of real acoustic data are discussed. The estimation of true geophysical parameters from acoustic observations often is an ill-conditioned problem that is regularized by enforcing prior constraints such as sparsity or energy penalities, and by reducing the size of the parameter search. Traditionally, empirical orthogonal functions (EOFs) and overcomplete wavelet and curvelet dictionaries have been used to represent complex geophysical structures with few parameters. Using the K-SVD dictionary learning algorithm, the representation of ocean SSP data is significantly compressed relative to EOF analysis. The regularization performance of these learned dictionaries is evaluated against EOFs in the estimation of ocean sound speed structure from ocean acoustic observations and the limitations of such unsupervised methods are considered.

A Chinese character structure preserved denoising method for Chinese tablet calligraphy document images based on KSVD dictionary learning

As an art form, Chinese ancient calligraphy tablet works occupy an important position in the heritage of Chinese culture. However, because of natural or man-made decay, there appear lots of noises in these ancient tablet works images, which have an important effect on the quality of the tablet images. To address this problem, a character structure preserved denoising method based on KSVD dictionary learning was proposed in this paper. This new proposed method consists of two major operations: dividing-frequency denoising and ant-like noises removal in binary image. At the stage of dividing-frequency denosing, the Butterworth low pass filter was employed to filter and extract low frequency part of the images firstly. Then, KSVD dictionary learning algorithm was used for smoothing the high frequency image and extracting image edges and the extracted edge images was then fused with the denoised low frequency part of the images. At the stage of ant-like noises removal, the fused image is converted into a binary one firstly. Then, the connected region method is employed to remove isolated ant-like noise; ergodic method is used to fill holes of strokes. Finally strokes thorn was eliminated by using the median filter. Experimental results demonstrate that the proposed method can effectively remove most image noise (including various block noise, linear noise and ant-like noise) and preserve characters better than existing methods.

An AK-BRP dictionary learning algorithm for video frame sparse representation in compressed sensing

Sparsifying transform is an important prerequisite in compressed sensing. And it is practically significant to research the fast and efficient signal sparse representation methods. In this paper, we propose an adaptive K-BRP (AK-BRP) dictionary learning algorithm. The bilateral random projection (BRP), a method of low rank approximation, is used to update the dictionary atoms. Furthermore, in the sparse coding stage, an adaptive sparsity constraint is utilized to obtain sparse representation coefficient and helps to improve the efficiency of the dictionary update stage further. Finally, for video frame sparse representation, our adaptive dictionary learning algorithm achieves better performance than K-SVD dictionary learning algorithm in terms of computation cost. And our method produces smaller reconstruction error as well.

Anti-interference Mechanism of ICA Blind Source Separation Combined with Wavelet De-noising

In the issues of signal de-noising, utilize K-SVD and other classical dictionary learning algorithm for sparse decomposition and reconstruction of signal, which cannot effectively eliminate influence of noise. The method suggested by this paper makes some improvement to the classical dictionary learning. Firstly, utilize K-SVD algorithm to make the dictionary learning; then, utilize the method of non-linear least squares to fit each atom in the dictionary and get the revised dictionary; finally, utilize the method of Particle Swarm Optimization to solve the spare representation of signal and get the reconstructed signal at last. It is proved through the experience that the de-noising effect of this paper is obvious superior to the conventional dictionary learning algorithm and is close to the effect of wavelet analytical approach.

Blind Separation of Permuted Alias Image Based on K-SVD Dictionary Learning

In this work, a new blind separation algorithm for permuted alias image based on dictionary learning is proposed according to a type of permuted alias image with noise diversity. Sparse representation of permuted alias image is obtained by dictionary learning method, since it has high adaptability and its representation result has higher sparsity degrees than that of parameter dictionary. An optimal permuted alias image is achieved by conducting sparse representation with K-SVD dictionary learning algorithm restrained with nonzero element number. The size and the location of permuting region is found by detecting the subtraction image, which is defined as the difference between the reconstructed permuted alias image and the original permuted alias image. The permuting region is optimized by implementing image morphological operation and is separated from the permuted alias image by the threshold. Experimental results show that the permuting sub-images can be efficiently separated from the permuted alias image, which is not affected by the size, location, number of permuting sub-images and noise level of the permuting sub-images.

ECG signal reconstruction from undersampled measurement using a trained overcomplete dictionary

We propose a new approach to reconstructing ECG signal from undersampled data based on constructing a combined overcomplete dictionary. The dictionary is obtained by combining the trained dictionary by K-SVD dictionary learning algorithm with universal types of dictionary such as DCT or wavelet basis. Using the trained overcomplete dictionary, the proposed method can find sparse approximation by compressive sensing. Experimental results on MIT-BIH arrhythmia database confirm that our proposed algorithm has high reconstruction performance while maintaining low distortion.

Dictionary learning based sparse coefficients for audio classification with max and average pooling

Audio classification is an important problem in signal processing and pattern recognition with potential applications in audio retrieval, documentation and scene analysis. Common to general signal classification systems, it involves both training and classification (or testing) stages. The performance of an audio classification system, such as its complexity and classification accuracy, depends highly on the choice of the signal features and the classifiers. Several features have been widely exploited in existing methods, such as the mel-frequency cepstrum coefficients (MFCCs), line spectral frequencies (LSF) and short time energy (STM). In this paper, instead of using these well-established features, we explore the potential of sparse features, derived from the dictionary of signal atoms using sparse coding based on e.g. orthogonal matching pursuit (OMP), where the atoms are adapted directly from audio training data using the K-SVD dictionary learning algorithm. To reduce the computational complexity, we propose to perform pooling and sampling operations on the sparse coefficients. Such operations also help to maintain a unified dimension of the signal features, regardless of the various lengths of the training and testing signals. Using the popular support vector machine (SVM) as the classifier, we examine the performance of the proposed classification system for two binary classification problems, namely speech-music classification and male-female speech discrimination and a multi-class problem, speaker identification. The experimental results show that the sparse (max-pooled and average-pooled) coefficients perform better than the classical MFCCs features, in particular, for noisy audio data.

Denoising infrared maritime imagery using tailored dictionaries via modified K-SVD algorithm

Recent work has shown that tailored overcomplete dictionaries can provide a better image model than standard basis functions for a variety of image processing tasks. Here we propose a modified K-SVD dictionary learning algorithm designed to maintain the advantages of the original approach but with a focus on improved convergence. We then use the learned model to denoise infrared maritime imagery and compare the performance to the original K-SVD algorithm, several overcomplete "fixed" dictionaries, and a standard wavelet denoising algorithm. Results indicate the superiority of overcomplete representations and show that our tailored approach provides similar peak signal-to-noise ratios as the traditional K-SVD at roughly half the computational cost.