Building Implicit Dictionaries Based on Extreme Random Clustering for Modality Recognition

The visual word dictionary is based on measuring metrics at the image level and does not follow a “decision-support tool” approach, where the system is used to aid diagnosis, as it is found in most of the current methods. The decision-support approach requires pinpointing the location of each lesion to allow the specialist to evaluate the image for diagnosis. In this case, metrics based on the accuracy of detecting each type of lesion is more relevant. Using a visual words dictionary as the basis for a classification algorithm was inspired by the computer vision, where visual dictionaries and PoIs are used as a basis for several applications such as image retrieval and classification. The visual word that combines feature information contained within the images in a framework easily extendible to different types of retinal lesions or pathologies and builds a specific projection space for each class of interest (e.g., white lesions such as exudates or normal regions) instead of a common dictionary for all classes. The visual words dictionary was applied to classifying bright and red lesions with classical cross validation and cross dataset validation to indicate the robustness of this approach. The curve (AUC) of 95.3% for white lesion detection and an AUC of 93.3% for red lesion detection using fivefold cross validation and data consisting of 687 images of normal retinae, 245 images with bright lesions, 191 with red lesions, and 109 with signs of both bright and red lesions. The image classification resulted in an AUC of 88.1% when classifying the RetiDB dataset and in an AUC of 89.3% when classifying the Messidor dataset, both cases for bright lesion detection. The results indicate the potential for training with different acquisition images under different setup conditions with a high accuracy of referral based on the presence of either red or bright lesions or both.

Characterizing images by high-level concepts from a learned visual dictionary is extensively used in image classification and retrieval. This paper deals with inferring discriminative visual dictionaries for effective image retrieval and examines a non-negative visual dictionary learning scheme towards this direction. More specifically, a non-negative matrix factorization framework with \(\ell _0\)-sparseness constraint on the coefficient matrix for optimizing the dictionary is proposed. It is a two-step iterative process composed of sparse encoding and dictionary enhancement stages. An initial estimate of the visual dictionary is updated in each iteration with the proposed \(\ell _0\)-constraint gradient projection algorithm. A desirable attribute of this formulation is an adaptive sequential dictionary initialization procedure. This leads to a sharp drop down of the approximation error and a faster convergence. Finally, the proposed dictionary optimization scheme is used to derive a compact image representation for the retrieval task. A new image signature is obtained by projecting local descriptors on to the basis elements of the optimized visual dictionary and then aggregating the resulting sparse encodings in to a single feature vector. Experimental results on various benchmark datasets show that the proposed system can infer enhanced visual dictionaries and the derived image feature vector can achieve better retrieval results as compared to state-of-the-art techniques.

Effective representation and classification of scenes using very high resolution (VHR) remote sensing images cover a wide range of applications. Although robust low‐level image features have been proven to be effective for scene classification, they are not semantically meaningful and thus have difficulty to deal with challenging visual recognition tasks. In this study, the authors propose a new and effective auto‐encoder‐based method to learn a shared mid‐level visual dictionary. This dictionary serves as a shared and universal basis to discover mid‐level visual elements. On the one hand, the mid‐level visual dictionary learnt using machine learning technique is more discriminative and contains rich semantic information, compared with the traditional low‐level visual words. On the other hand, the mid‐level visual dictionary is more robust to occlusions and image clutters. In the authors' scene‐classification scheme, they use discriminative mid‐level visual elements, rather than individual pixels or low‐level image features, to represent images. This new image representation is able to capture much of the high‐level meaning and contents of the image, facilitating challenging remote sensing image scene‐classification tasks. Comprehensive evaluations on a challenging VHR remote sensing images data set and comparisons with state‐of‐the‐art approaches demonstrate the effectiveness and superiority of their study.

This paper proposes an adaptive image representation learning method for cervix cancer tumor detection. The method learns the representation in two stages, a local feature description using a sparse dictionary learning and a global image representation using a bag-of-features (BOF) approach. The resultant representation is thus a BOF histogram, learned from a sparse local patch representation. The parameters of the sparse learning representation algorithm are tuned up by searching dictionaries with low coherence and high sparsity. The proposed method was evaluated in a dataset of 394 cervical histology images with tumoral and non-tumoral pathologies acquired at a 10X magnification and a resolution of 3800 × 3000 pixels in RGB color. A conventional BOF image representation, using a linearized raw-block patch descriptor, was selected as the baseline. The preliminary results show that our proposed method improves the baseline for all different BOF dictionary sizes (125, 250, 500, 1000 and 2000). Under a 10 cross-validation test and a 2000 BOF dictionary, the best performance was 0:77±0:04 in average accuracy, improving in 2:5% the baseline. These results suggest that a learning-from-data approach could be used in different stages of an image classifier construction pipeline, in particular for the image representation stage.

Dictionary learning of spectrograms consists of detecting their fundamental spectra-temporal patterns and their associated activation signals. In this paper, we propose an efficient convolutive dictionary learning approach for analyzing repetitive bioacoustics patterns from a collection of audio recordings. Our method is inspired by the convolutive non-negative matrix factorization (CNMF) model. The proposed approach relies on random projection for reduced computational complexity. As a consequence, the non-negativity requirement on the dictionary words is relaxed. Moreover, the proposed approach is well-suited for a collection of discontinuous spectrograms. We evaluate our approach on synthetic examples and on two real datasets consisting of multiple birds audio recordings. Bird syllable dictionary learning from a real-world dataset is demonstrated. Additionally, we apply the approach for spectrogram denoising in the presence of rain noise artifacts.

An important issue in statistical modeling is to determine the complexity of the model based on the scale of data so as to effectively mitigate the model’s overfitting problems without big data. We adopt a data-driven approach to automatically determine the number of components of the model. In order to better extract robust features, we propose a framework of data-driven two-layer structure visual dictionary learning (DTSVDL). It works by dividing the visual dictionary structure learning into two levels: the attribute layer and the detail layer. In the attribute layer, the attributes of the image dataset are learned, and these attributes are obtained by a data-driven Bayesian nonparametric model. Then, in the detail layer, the detailed information over attributes is further explored and refined, and the attributes are weighted by the number of effective observations associated with each attribute. Our proposed approach has three main advantages: (1) the two-layer structure makes our building visual dictionary be more expressive; (2) the number of components in the attribute layer can be determined automatically from the data; (3) the components are automatically determined based on the scale of visual words; therefore, our model can well mitigate the overfitting problem. In addition, by comparing with stacked autoencoders, stacked denoising autoencoders, LeNet-5, speeded-up robust features, and pretrained deep learning model ImageNet-VGG-F algorithms, we find that our approach achieves satisfactory image categorization results on two benchmark datasets. Specifically, higher categorization performance is achieved than by the classical approaches on 15 scene categories and action datasets. We conclude that the resulting DTSVDL possesses a good generality derived from attribute information as well as an excellent distinction derived from detailed information. In other words, the visual dictionary learned by our algorithm is more expressive and discriminatory.

The ever increasing number of medical images in hospitals urges on the need for generic content based image retrieval systems. These systems are in an area of great importance to the healthcare providers. The first and foremost function in such system is feature extraction. In this paper, different feature extraction techniques have been utilized to represent medical blood cell images. They are categorized into two groups; low-level image representation such as color and shape analysis and local patch-based image representation such as Bag of Words (BoW). These features have been exploited for retrieving similar images. We have also used a generative model such as Probabilistic Latent Semantic Analysis (PLSA) on extracted BoW for retrieval task. Lastly, the retrieval results obtained from all the above features are integrated with one another to increase the retrieval performance. Experimental results using four different classes of 600 blood cell images showed 92.25% of retrieval accuracy.

Dictionary learning has been widely used in image representation under the framework of sparse theory. But most of the current dictionary learning strategies can only be used for single-frame image separately, which are insufficient from the perspective of incremental information acquisition and global optimization for the sequential or multiframe satellite images. To this end, this paper proposes an incremental dictionary learning method for multiframe satellite images representation in the spectral domain. The incremental dictionary learning is formulated analytically in the framework of sparse representation with low-rank constraint, as a frame-by-frame gradual optimization process of global and local dictionaries, and their corresponding sparse coefficients with the sequence. Specifically, the global dictionary representing the common spectral information of the sequential frames, is optimized by two adjacent frames gradually. Meanwhile, the local dictionary representing the specific spectral information of each frame, is optimized by the newly added frame itself. In addition, an activity ratio for separating the global dictionary from the local dictionaries, an outlier detection method for initializing the local dictionary are also given, and the alternating direction methods of multipliers (ADMM) is employed to implement the above optimization. Comparison results with the related state-of-the-art methods on different datasets demonstrate that, our proposed method achieves the best representation performance in both spatial and spectral domains, and also helps to improve the performance of dictionary-based tasks using sequential satellite images, such as sea surface anomaly detection.

Sparse representation is efficient to approximately recover signals by a linear composition of a few bases from an over-complete dictionary. However, in the scenario of data compression, its efficiency and popularity are hindered due to the extra overhead for encoding the sparse coefficients. Therefore, how to establish an accurate rate model in sparse coding and dictionary learning becomes meaningful, which has been not fully exploited in the context of sparse representation. According to the Shannon entropy inequality, the variance of data source can bound its entropy, thus can reflect the actual coding bits. Therefore, a Globally Variance-Constrained Sparse Representation (GVCSR) model is proposed, where a variance-constrained rate term is introduced to the conventional sparse representation. To solve the non-convex optimization problem, we employ the Alternating Direction Method of Multipliers (ADMM) for sparse coding and dictionary learning, both of which have shown state-of-the-art rate-distortion performance in image representation.

This paper analyzes the role of dictionary selection in Sparse Representation-based Classification (SRC). While SRC introduces interesting results in the field of classification, its performance is highly limited by the number of training samples to form the classification matrix. Different studies addressed this issue by using a more compact representation of the training data in order to achieve higher classification speed and accuracy. Representative selection methods which are analyzed in this paper include Metaface dictionary learning, Fisher Discriminative Dictionary Learning (FDDL), Sparse Modeling Representative Selection (SMRS), and random selection of the training samples. The first two methods build their own dictionaries via an optimization process while the other two methods select the representatives directly from the original training samples. These methods, along with the original method which uses all training samples to form the classification matrix, were examined on two face datasets and one digit dataset. The role of feature extraction was also studied using two dimensionality reduction methods, down-sampling and random projection. The results show that the FDDL method leads to the best classification accuracy followed by the SMRS method as the second best. On the other hand, the SMRS method requires a much smaller learning time which makes it more appropriate for dynamic situations where the dictionary is regularly updated with new samples. The accuracy of the Metaface dictionary learning method was specifically less than the other two methods. As expected, using all the training samples as the dictionary resulted in the best recognition rates in all the datasets but the classification times for this approach were far larger than the required time using any of the three dictionary learning methods.

Sample diversity, representation effectiveness and robust dictionary learning for face recognition

Image representations, from SIFT and Bag of Visual Words to Convolutional Neural Networks (CNNs), are a crucial component of almost any image understanding system. Nevertheless, our understanding of them remains limited. In this paper we conduct a direct analysis of the visual information contained in representations by asking the following question: given an encoding of an image, to which extent is it possible to reconstruct the image itself? To answer this question we contribute a general framework to invert representations. We show that this method can invert representations such as HOG more accurately than recent alternatives while being applicable to CNNs too. We then use this technique to study the inverse of recent state-of-the-art CNN image representations for the first time. Among our findings, we show that several layers in CNNs retain photographically accurate information about the image, with different degrees of geometric and photometric invariance.

K-Singular Value Decomposition (K-SVD) is an excellent feature analysis method, but it has a large amount of calculation and low algorithm efficiency. To solve the above problems, a dictionary learning method for image block classification training is proposed. This method uses image entropy as the classification index of image samples, selects more complex texture image blocks with more analytical value, and uses Discrete Cosine Transform- Discrete Wavelet Transformation (DCT-DWT) dictionary to train them in K-Singular Value Decomposition. Through the feedback of sparse coefficients, the threshold value is dynamically changed and the precision of sparse coding is controlled. In the sparse solution, a variable-step Orthogonal Matching Pursuit (OMP) algorithm is used. The experimental comparison shows that the classification K-Singular Value Decomposition dictionary learning algorithm runs faster, and the Peak signal-to-noise Ratio (PSNR) and Structural Similarity (SSIM) value of the reconstructed image are higher.

In this paper, we propose a multiscale dictionary learning framework for hierarchical sparse representation of natural images. The proposed framework leverages an adaptive quadtree decomposition to represent structured sparsity in different scales. In dictionary learning, a tree-structured regularized optimization is formulated to distinguish and represent high-frequency details based on varying local statistics and group low-frequency components for local smoothness and structural consistency. In comparison to traditional proximal gradient method, block-coordinate descent is adopted to improve the efficiency of dictionary learning with a guarantee of recovery performance. The proposed framework enables hierarchical sparse representation by naturally organizing the trained dictionary atoms in a prespecified arborescent structure with descending scales from root to leaves. Consequently, the approximation of high-frequency details can be improved with progressive refinement from coarser to finer scales. Employed into image denoising, the proposed framework is demonstrated to be competitive with the state-of-the-art methods in terms of objective and visual restoration quality.

The bag of visual words (BOW) model is an efficient image representation technique for image categorization and annotation tasks. Building good visual vocabularies, from automatically extracted image feature vectors, produces discriminative visual words, which can improve the accuracy of image categorization tasks. Most approaches that use the BOW model in categorizing images ignore useful information that can be obtained from image classes to build visual vocabularies. Moreover, most BOW models use intensity features extracted from local regions and disregard colour information, which is an important characteristic of any natural scene image. In this paper, we show that integrating visual vocabularies generated from each image category improves the BOW image representation and improves accuracy in natural scene image classification. We use a keypoint density-based weighting method to combine the BOW representation with image colour information on a spatial pyramid layout. In addition, we show that visual vocabularies generated from training images of one scene image dataset can plausibly represent another scene image dataset on the same domain. This helps in reducing time and effort needed to build new visual vocabularies. The proposed approach is evaluated over three well-known scene classification datasets with 6, 8 and 15 scene categories, respectively, using 10-fold cross-validation. The experimental results, using support vector machines with histogram intersection kernel, show that the proposed approach outperforms baseline methods such as Gist features, rgbSIFT features and different configurations of the BOW model.