Initial Dictionary Research Articles

An improved collaborative representation based classification with regularized least square (CRC–RLS) method for robust face recognition

Fast and robust face recognition is a challenging research topic in the field of computer vision. A recently proposed Collaborative Representation based Classification with Regularized Least Square (CRC–RLS) algorithm shows very lower computational cost but with poor robustness. In order to solve this problem, we propose an improved CRC–RLS method. Firstly, the image Gabor features were extracted and used to construct initial dictionary. Secondly, we learn a discriminative dictionary by a label consistent K-SVD (LC-KSVD) method which combines the sparse coding error with the reconstruction error and the classification error. Finally, l2-norm of coding residual in CRC–RLS is computed and the classification problem is transformed into solving linear programing problem. Experiments on two benchmark face databases with variations of illumination, expression, occlusion show that the proposed method can achieve high classification accuracy and has a very low time-consuming.

Segmentation of melanocytic skin lesions using feature learning and dictionaries

Pre-screening systems for the diagnosis of melanocytic skin lesions depend of the proper segmentation of the image region affected by the lesion. This paper proposes a feature learning scheme that finds relevant features for skin lesion image segmentation. This work introduces a new unsupervised dictionary learning method, namely Unsupervised Information-Theoretic Dictionary Learning (UITDL), and discusses how it can be applied in the segmentation of skin lesions in macroscopic images. The UITDL approach is adaptive and tends to be robust to outliers in the training data, and consists of two main stages. In the first stage, a textural variation image is used to construct an initial feature dictionary and an initial sparse representation via Non-Negative Matrix Factorization (NMF). In the second stage, the feature dictionary is optimized by selecting adaptively the number of dictionary atoms. The greedy approach used for dictionary optimization is quite efficient and flexible enough to be applied to other dictionary learning problems. Furthermore, the proposed method can be easily extended for other image segmentation problems. The experimental results suggest that the proposed approach potentially can provide more accurate skin lesion segmentation results than comparable state-of-the-art methods. The proposed segmentation method could help to improve the performance of pre-screening systems for melanocytic skin lesions, which can affect positively the quality of the early diagnosis provided to skin lesion patients.

Sparse representation-based classification using generalized weighted extended dictionary

Sparse representation-based classification (SRC) has been a breakthrough of face recognition and signal reconstruction recently. However, few face images from the same subject provide insufficient observations. Meanwhile, owing to uncertainty of training images with variations in the appearance of facial illumination, pose and facial expression, it is difficult to successfully address the issues of image recognition and signal reconstruction. Thus, how to construct an optimized comprehensive dictionary is still a non-trivial task. In this paper, we develop a generalized virtual extended dictionary of SRC (V-SRC) to deal with data uncertainty problem, establishing a weighted mechanism which assigns different coefficients between arbitrary two within-class samples to obtain the virtual samples. More specifically, we firstly construct a set of synthesized training samples by means of weighted combination of existing training samples. Secondly, we combine the original and synthesized training samples as an initial extended dictionary. Thirdly, in order to decrease the constructed error in sparse representation, we exploit an elimination scheme to gain the robust training samples with respect to the initial extended training dictionary. The final goal of the proposed method is to utilize the linear combination of competitive training samples to perform face classification. Experimental results conducted on the AR, FERET and Georgia Tech databases demonstrate the effectiveness of the proposed method especially in the case of small sample size problem.

Three-dimensional segmentation of retinal cysts from spectral-domain optical coherence tomography images by the use of three-dimensional curvelet based K-SVD

This paper presents a new three-dimensional curvelet transform based dictionary learning for automatic segmentation of intraretinal cysts, most relevant prognostic biomarker in neovascular age-related macular degeneration, from 3D spectral-domain optical coherence tomography (SD-OCT) images. In particular, we focus on the Spectralis SD-OCT (Heidelberg Engineering, Heidelberg, Germany) system, and show the applicability of our algorithm in the segmentation of these features. For this purpose, we use recursive Gaussian filter and approximate the corrupted pixels from its surrounding, then in order to enhance the cystoid dark space regions and future noise suppression we introduce a new scheme in dictionary learning and take curvelet transform of filtered image then denoise and modify each noisy coefficients matrix in each scale with predefined initial 3D sparse dictionary. Dark pixels between retinal pigment epithelium and nerve fiber layer that were extracted with graph theory are considered as cystoid spaces. The average dice coefficient for the segmentation of cystoid regions in whole 3D volume and with-in central 3 mm diameter on the MICCAI 2015 OPTIMA Cyst Segmentation Challenge dataset were found to be 0.65 and 0.77, respectively.

A fast and robust face recognition approach combining Gabor learned dictionaries and collaborative representation

Proposed is a simple yet fast and robust approach to face recognition. This approach is developed specifically to address the challenges due to variations of illumination, expression and occlusion, when studying the facial images of a large population. The proposed approach exploits an improved collaborative representation algorithm. First, we construct an initial dictionary by extracting the multi-scale and multi-orientation Gabor features of the image. Second, we design a new discriminative dictionary by an improved K-SVD algorithm, so that the query sample can be better represented for classification. Finally, l2-norm of coding residual is calculated by collaborative representation based classification with regularized least square method (CRC-RLS) and the class of test sample is obtained. Experiments on two benchmark face datasets show that the proposed method can achieve high classification accuracy and is comparatively low in terms of time-consumption, compared to the CRC-RLS algorithm.

Enhancement of snow cover change detection with sparse representation and dictionary learning

Abstract. Sparse representation and decoding is often used for denoising images and compression of images with respect to inherent features. In this paper, we adopt a methodology incorporating sparse representation of a snow cover change map using the K-SVD trained dictionary and sparse decoding to enhance the change map. The pixels often falsely characterized as "changes" are eliminated using this approach. The preliminary change map was generated using differenced NDSI or S3 maps in case of Resourcesat-2 and Landsat 8 OLI imagery respectively. These maps are extracted into patches for compressed sensing using Discrete Cosine Transform (DCT) to generate an initial dictionary which is trained by the K-SVD approach. The trained dictionary is used for sparse coding of the change map using the Orthogonal Matching Pursuit (OMP) algorithm. The reconstructed change map incorporates a greater degree of smoothing and represents the features (snow cover changes) with better accuracy. The enhanced change map is segmented using kmeans to discriminate between the changed and non-changed pixels. The segmented enhanced change map is compared, firstly with the difference of Support Vector Machine (SVM) classified NDSI maps and secondly with a reference data generated as a mask by visual interpretation of the two input images. The methodology is evaluated using multi-spectral datasets from Resourcesat-2 and Landsat-8. The k-hat statistic is computed to determine the accuracy of the proposed approach.

Information-Theoretic Dictionary Learning for Image Classification

We present a two-stage approach for learning dictionaries for object classification tasks based on the principle of information maximization. The proposed method seeks a dictionary that is compact, discriminative, and generative. In the first stage, dictionary atoms are selected from an initial dictionary by maximizing the mutual information measure on dictionary compactness, discrimination and reconstruction. In the second stage, the selected dictionary atoms are updated for improved reconstructive and discriminative power using a simple gradient ascent algorithm on mutual information. Experiments using real data sets demonstrate the effectiveness of our approach for image classification tasks.

Emergency Medical Text Classifier: New system improves processing and classification of triage notes.

ObjectiveAutomated syndrome classification aims to aid near real-time syndromic surveillance to serve as an early warning system for disease outbreaks, using Emergency Department (ED) data. We present a system that improves the automatic classification of an ED record with triage note into one or more syndrome categories using the vector space model coupled with a ‘learning’ module that employs a pseudo-relevance feedback mechanism. Materials and Methods: Terms from standard syndrome definitions are used to construct an initial reference dictionary for generating the syndrome and triage note vectors. Based on cosine similarity between the vectors, each record is classified into a syndrome category. We then take terms from the top-ranked records that belong to the syndrome of interest as feedback. These terms are added to the reference dictionary and the process is repeated to determine the final classification. The system was tested on two different datasets for each of three syndromes: Gastro-Intestinal (GI), Respiratory (Resp) and Fever-Rash (FR). Performance was measured in terms of sensitivity (Se) and specificity (Sp). Results: The use of relevance feedback produced high values of sensitivity and specificity for all three syndromes in both test sets: GI: 90% and 71%, Resp: 97% and 73%, FR: 100% and 87%, respectively, in test set 1, and GI: 88% and 69%, Resp: 87% and 61%, FR: 97% and 71%, respectively, in test set 2. Conclusions: The new system for pre-processing and syndromic classification of ED records with triage notes achieved improvements in Se and Sp. Our results also demonstrate that the system can be tuned to achieve different levels of performance based on user requirements.

CSMMI: class-specific maximization of mutual information for action and gesture recognition.

In this paper, we propose a novel approach called class-specific maximization of mutual information (CSMMI) using a submodular method, which aims at learning a compact and discriminative dictionary for each class. Unlike traditional dictionary-based algorithms, which typically learn a shared dictionary for all of the classes, we unify the intraclass and interclass mutual information (MI) into an single objective function to optimize class-specific dictionary. The objective function has two aims: 1) maximizing the MI between dictionary items within a specific class (intrinsic structure) and 2) minimizing the MI between the dictionary items in a given class and those of the other classes (extrinsic structure). We significantly reduce the computational complexity of CSMMI by introducing an novel submodular method, which is one of the important contributions of this paper. This paper also contributes a state-of-the-art end-to-end system for action and gesture recognition incorporating CSMMI, with feature extraction, learning initial dictionary per each class by sparse coding, CSMMI via submodularity, and classification based on reconstruction errors. We performed extensive experiments on synthetic data and eight benchmark data sets. Our experimental results show that CSMMI outperforms shared dictionary methods and that our end-to-end system is competitive with other state-of-the-art approaches.

Sparse Representation Based Pansharpening Using Trained Dictionary

Sparse representation has been used to fuse high-resolution panchromatic (HRP) and low-resolution multispectral (LRM) images. However, the approach faces the difficulty that the dictionary is generated from the high-resolution multispectral (HRM) images, which are unknown. In this letter, a two-step method is proposed to train the dictionary from the HRP and LRM images. In the first step, coarse HRM images are obtained by additive wavelet fusion method. The initial dictionary is composed of randomly sampled patches from the coarse HRM images. In the second step, a linear constraint K-SVD method is designed to train the dictionary to improve its representation ability. Experimental results using QuickBird and IKONOS data indicate that the trained dictionary yields comparable fusion products with raw-patch-dictionary sampled from HRM images.

Learning Dictionaries With Bounded Self-Coherence

Sparse coding in learned dictionaries has been established as a successful approach for signal denoising, source separation and solving inverse problems in general. A dictionary learning method adapts an initial dictionary to a particular signal class by iteratively computing an approximate factorization of a training data matrix into a dictionary and a sparse coding matrix. The learned dictionary is characterized by two properties: the coherence of the dictionary to observations of the signal class, and the self-coherence of the dictionary atoms. A high coherence to the signal class enables the sparse coding of signal observations with a small approximation error, while a low self-coherence of the atoms guarantees atom recovery and a more rapid residual error decay rate for the sparse coding algorithm. The two goals of high signal coherence and low self-coherence are typically in conflict, therefore one seeks a trade-off between them, depending on the application. We present a dictionary learning method with an effective control over the self-coherence of the trained dictionary, enabling a trade-off between maximizing the sparsity of codings and approximating an equi-angular tight frame.

An augmented Lagrangian approach to general dictionary learning for image denoising

This paper presents an augmented Lagrangian (AL) based method for designing of overcomplete dictionaries for sparse representation with general lq-data fidelity term (q⩽2). In the proposed method, the dictionary is updated via a simple gradient descent method after each inner minimization step of the AL scheme. Besides, a modified Iterated Shrinkage/Thresholding Algorithm is employed to accelerate the sparse coding stage of the algorithm. We reveal that the dictionary update strategy of the proposed method is different from most of existing methods because the learned dictionaries become more and more complex regularly. An advantage of the iterated refinement methodology is that it makes the method less dependent on the initial dictionary. Experimental results on real image for Gaussian noise removal (q=2) and impulse noise removal (q=1) consistently demonstrate that the proposed approach can efficiently remove the noise while maintaining high image quality.

Adaptive matching pursuit with constrained total least squares

Abstract Compressive sensing (CS) can effectively recover a signal when it is sparse in some discrete atoms. However, in some applications, signals are sparse in a continuous parameter space, e.g., frequency space, rather than discrete atoms. Usually, we divide the continuous parameter into finite discrete grid points and build a dictionary from these grid points. However, the actual targets may not exactly lie on the grid points no matter how densely the parameter is grided, which introduces mismatch between the predefined dictionary and the actual one. In this article, a novel method, namely adaptive matching pursuit with constrained total least squares (AMP-CTLS), is proposed to find actual atoms even if they are not included in the initial dictionary. In AMP-CTLS, the grid and the dictionary are adaptively updated to better agree with measurements. The convergence of the algorithm is discussed, and numerical experiments demonstrate the advantages of AMP-CTLS.

An augmented Lagrangian multi-scale dictionary learning algorithm

AbstractLearning overcomplete dictionaries for sparse signal representation has become a hot topic fascinated by many researchers in the recent years, while most of the existing approaches have a serious problem that they always lead to local minima. In this article, we present a novel augmented Lagrangian multi-scale dictionary learning algorithm (ALM-DL), which is achieved by first recasting the constrained dictionary learning problem into an AL scheme, and then updating the dictionary after each inner iteration of the scheme during which majorization-minimization technique is employed for solving the inner subproblem. Refining the dictionary from low scale to high makes the proposed method less dependent on the initial dictionary hence avoiding local optima. Numerical tests for synthetic data and denoising applications on real images demonstrate the superior performance of the proposed approach.

Recursive Least Squares Dictionary Learning Algorithm

We present the recursive least squares dictionary learning algorithm, RLS-DLA, which can be used for learning overcomplete dictionaries for sparse signal representation. Most DLAs presented earlier, for example ILS-DLA and K-SVD, update the dictionary after a batch of training vectors has been processed, usually using the whole set of training vectors as one batch. The training set is used iteratively to gradually improve the dictionary. The approach in RLS-DLA is a continuous update of the dictionary as each training vector is being processed. The core of the algorithm is compact and can be effectively implemented. The algorithm is derived very much along the same path as the recursive least squares (RLS) algorithm for adaptive filtering. Thus, as in RLS, a forgetting factor ? can be introduced and easily implemented in the algorithm. Adjusting ? in an appropriate way makes the algorithm less dependent on the initial dictionary and it improves both convergence properties of RLS-DLA as well as the representation ability of the resulting dictionary. Two sets of experiments are done to test different methods for learning dictionaries. The goal of the first set is to explore some basic properties of the algorithm in a simple setup, and for the second set it is the reconstruction of a true underlying dictionary. The first experiment confirms the conjectural properties from the derivation part, while the second demonstrates excellent performance.

Matching Pursuit-Based Region-of-Interest Image Coding

Matching pursuit (MP) is a multiresolution signal analysis method and can be used to render a selected region of an image with a specific quality. A novel, scalable, and progressive MP-based region-of-interest image-coding scheme is presented. The method is capable of providing a trade off between rate, distortion, and complexity. The method also provides an interactive way of information refinement for regions of an image with higher receiver's priority. By selecting a proper subset of the huge initial MP dictionary, using the method described in this paper, the complexity burden of MP analysis can be adapted to the computational power of the image encoder.

A dictionary and self defining protocol for exchanging power system information

The Power System Analytical Data Task Force has been meeting since the early 1990s with the objective of developing a data dictionary containing definitions of power system analytical data. A dictionary template has been adopted and work on an initial dictionary is underway. A description of the template and the proposed draft dictionary is the subject of this paper. This paper also proposes a protocol for exchanging information contained in a database that has a defined dictionary. Using a typical power system example, a dictionary is defined and the use of the proposed protocol is demonstrated.

劣化推定に基づいた辞書の自動選択による本棚画像中の文字認識

We developed an adaptive dictionary-selection method for recognizing low-quality characters in scene images. First, the environmental conditions of the input image are estimated using an initial dictionary. Then, a relevant dictionary from several dictionaries reflecting different environmental conditions is automatically selected based on the estimation and used for recognition. Experiments on characters in images of bookshelves showed that this method achieves a higher recognition rate (89.8%) than that using one fixed dictionary (76.4%). Furthermore, recognition accuracy improved from 89.8% to 95.2% when contextual postprocessing was used.

Dynamic dictionary matching with failure functions

Amir and Farach (1991) and Amir et al. (to appear) recently initiated the study of the dynamic dictionary pattern matching problem. The dictionary D contains a set of patterns that can change over time by insertion and deletion of individual patterns. The user may also present a text string and ask to search for all occurrences of any patterns in the text. For the static dictionary problem, Aho and Corasick (1975) gave a strategy based on a failure function automaton that takes O(| D|log|Σ|) time to build a dictionary of size | D| and searches a text T in time O(| T|log|Σ|+ tocc), where tocc, is the total number of pattern occurrences in the text. Amir et al. (to appear) used an automaton based on suffix trees to solve the dynamic problem. Their method can insert or delete a pattern P in time O(| P|log| D|) and can search a text in time O((| T|+ tocc)log| D|). We show that the same bounds can be achieved using a framework based on failure functions. We then show that our approach also allows us to achieve faster search times at the expense of the update times; for constant k, we can achieve linear O(| T|( k+log|Σ|)+ k tocc) search time with an update time of O( k| P∥ D| 1 k ). This is advantageous if the search texts are much larger than the dictionary or searches are more frequent than updates. Finally, we show how to build the initial dictionary in O(| D|log|Σ|) time, regardless of what combination of search and update times is used.