Search

• GML (Group Meta-Learning) is proposed to effectively alleviate the problem that low-quality samples have unfavourable impact on the training effect under few-shot conditions, thereby improving the performance of the model. • Momentum update strategy is introduced to few-shot learning for the first time to effectively improve the generalization and stability of the model, and an adaptive momentum coefficient is designed to form AMS (Adaptive Momentum Smoothing) to further improve the training efficiency. • We propose MGML (Momentum Group Meta-Learning) by combining GML and AMS. MGML not only improves the accuracy of Meta-Learning Baseline, but also has good mobility. It can be inserted into previous state-of-the-art methods and get consistent performance improvements. At present, image classification covers more and more fields, and it is often difficult to obtain enough data for learning in some specific scenarios, such as medical fields, personalized customization of robots, etc. Few-shot image classification aims to quickly learn new classes of features from few images, and the meta-learning method has become the mainstream due to its good performance. However, the generalization ability of the meta-learning method is still poor and easy to be disturbed by low-quality images. In order to solve the above problems, this paper proposes Momentum Group Meta-Learning (MGML) to achieve a better effect of few-shot learning, which contains Group Meta-Learning module (GML) and Adaptive Momentum Smoothing module (AMS). GML obtains an ensemble model by training multiple episodes in parallel and then grouping them, which can reduce the interference of low-quality samples and improve the stability of meta-learning training. AMS includes the adaptive momentum update rule to further optimally integrate models between different groups, so that the model can memorize experience in more scenarios and enhance the generalization ability. We conduct experiments on miniImageNet and tieredImageNet datasets. The results show that MGML improves the accuracy, stability and cross-domain transfer ability of few-shot image classification, and can be applied to different few-shot learning models. [ABSTRACT FROM AUTHOR]

Hashing technology can support large-scale recommendation very effectively due to its advantages of low storage cost and high recommendation efficiency. However, existing hashing-based recommendation methods often suffer from item cold-start problem. This is because they simply consider the user-item interaction and the single content information of the items, but the full interaction history is not always available and the single auxiliary information may be missing. To solve this issue, in this paper we propose a Binary Multi-modal Matrix Factorization (BMMF) method. First, we propose an efficient consensus multi-modal mapping to transform the heterogeneous multi-modal features to the unified factors by exploiting the complementarity of multiple modalities. Then, binary matrix factorization is simultaneously performed on the multi-modal features of the items and past user preferences to learn the compact binary codes of the users/items in a common Hamming space. In addition, inspired by the observation that similar instances often have similar binary codes within a short Hamming distance, we formulate a semantic structure regularization term to preserve the similarities of the items during the binary embedding process. Finally, we develop an effective Discrete Coordinate Descent (DCD) approach to tackle the formulated discrete hash optimization problem directly. Experiments on three publicly available real-world datasets demonstrate the superiority of the proposed method against the state-of-the-art methods. Our source codes and testing datasets are available at https://github.com/pcm1217/BMMF. [ABSTRACT FROM AUTHOR]

Cross-modal hashing (CMH) has gained considerable attention due to its low storage and fast query speed. Most existing CMH research pursues the binary codes of each modality by preserving a similarity matrix which is computationally expensive. In this paper, we propose a Gaussian similarity preserving method for cross-modal hashing. By using the semantic transformation, our model can avoid computing the similarity matrix explicitly. We further employ the sequential learning approach to reduce the quantization error. Experimental results on three benchmark datasets clearly show that our proposed method can outperform the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Searching for the nearest neighbor is a fundamental problem in the computer vision field, and deep hashing has become one of the most representative and widely used methods, which learns to generate compact binary codes for visual data. In this paper, we first delve into the representation learning of deep hashing and surprisingly find that deep hashing could be a double-edged sword, i.e. , deep hashing can accelerate the query speed and decrease the storage cost in the nearest neighbor search progress, but it greatly sacrifices the discriminability of deep representations especially with extremely short target code lengths. To solve this problem, we propose a two-step deep hashing learning framework. The first step focuses on learning deep discriminative representations with metric learning. Subsequently, the learning framework concentrates on simultaneously learning compact binary codes and preserving representations learned in the former step from being sacrificed. Extensive experiments on two general image datasets and four challenging image datasets validate the effectiveness of our proposed learning framework. Moreover, the side effect of deep hashing is successfully mitigated with our learning framework. [ABSTRACT FROM AUTHOR]

In this paper, we proposed a more robust supervised hashing framework based on the Cauchy loss function and Supervised Discrete Hashing (SDH) called Robust Supervised Discrete Hashing (RSDH), which can learn a robust subspace consisted of binary codes. The Cauchy loss is used to measure the error between the label matrix and the product of the decomposed matrices. RSDH can not only reduce the outliers and noise of the hashing codes, but also achieve the more satisfactory retrieval effect. Image retrieval experiments demonstrate that RSDH performs better than the other hashing methods. [ABSTRACT FROM AUTHOR]

There has been continuously growing research interest in Artificial Neural Networks (ANNs) due to their vast acceptance in many real-world applications. This has led to the appearance of a variety of realization methods for multilayer perceptron (MLP) inference models. In this paper, we propose a generalized 3-stage, configurable, multiplier-less, massive parallel realization architecture for MLP inference models. We use offset binary coded distributive arithmetic (OBC-DA), which uses LUTs to reduce hardware requirements, to realize internal computing units of MLPs. The designing efforts to achieve a reduction in resource requirements are two-fold; (i) reduction in LUT size itself, which is achieved by exploiting symmetry induced by OBC-DA and thereby reducing LUT size by 50%, and (ii) reduction in total number of LUTs which is attained by sharing the LUTs. The resource efficiency of the proposed architecture is further improved by decomposing the LUTs into parallel and smaller LUTs. Moreover, the resource efficiency is also enhanced by utilizing the symmetry among the inputs in the MLP layer. The ASIC synthesis results for underlying MLPs establish that the proposed method for the realization of MLPs is much more efficient than the earlier reported methods and coincides with the findings of the generalized hardware-timing complexities. [ABSTRACT FROM AUTHOR]

Multivariate time series forecasting has typically been a relevant and interesting topic in many fields, including economics, electricity consumption, solar energy, and traffic management. In these domains, owing to the complex dependencies among multiple variables and the mixed dependencies in the time dimension, it is challenging to forecast a multivariate time series precisely. Furthermore, most of the forecasting methods fail to capture the mixed influence of the different time-length dependencies among multiple variables. In this paper, a new deep learning framework is proposed for dealing with this challenging problem, named as mixed dependence time-series network (MDTNet). In this framework, stacked dilated convolutions and recurrent units are applied to extract the complex patterns in the long- and short-term mixed dependencies among multiple variables. The experiments show that our proposed framework yields significant results, outperforming the state-of-the-art baseline methods on three of the four benchmark datasets in large horizons and achieving a competitive performance in short horizons on all the benchmark datasets. [ABSTRACT FROM AUTHOR]

In this paper, a novel unsupervised C ross- M odal retrieval method via S ufficient S cene R epresentation (CMSSR) is proposed. Distinguished from the existing methods which mainly focus on simultaneously preserving the mutually-constrained intra- and inter-modal similarity relation, CMSSR considers data of different modalities as the descriptions of a scene from different views and accordingly integrates information of different modalities to learn a complete common representation containing sufficient information of the corresponding scene. To obtain such common representation, Gaussian Mixture Model (GMM) is firstly utilized to generate statistic representation of each uni-modal data, while the uni-modal spaces are accordingly abstracted as uni-modal statistical manifolds. In addition, the common space is assumed to be a high-dimensional statistical manifold with different uni-modal statistical manifolds as its sub-manifolds. In order to generate sufficient scene representation from uni-modal data, a representation completion strategy based on logistic regression is proposed to effectively complete the missing representation of another modality. Then, the similarity between different multi-modal data can be more accurately reflected by the distance metric in common statistical manifold. Based on the distance metric in common statistical manifold, Iterative Quantization is utilized to further generate binary code for fast cross-modal retrieval. Extensive experiments on three standard benchmark datasets fully demonstrate the superiority of CMSSR compared with several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

With the advent of the big data era, multimedia data is growing rapidly, and its data modalities is also becoming diversified. Therefore, the demand for the speed and accuracy of cross-modal information retrieval is increasing. Hashing-based cross-modal retrieval technology attracts widespread attention, it encodes multimedia data into a common binary hash space, thereby effectively measuring the correlation between samples from different modalities. In this paper, we propose a novel end-to-end deep cross-modal retrieval framework, namely Clustering-driven Deep Adversarial Hashing (CDAH), which has three main characteristics. Firstly, CDAH learns discriminative clusters recursively through a soft clustering model. It attempts to generate modal-invariant representations in a common space by obfuscating the modality classifier, which tries to distinguish different modalities according to the generated representations. Secondly, in order to minimize the modal gap between feature representations from different modalities with the same semantic label, and to maximize the distance between images and texts with different labels, CDAH constructs a fused-semantics matrix to integrate the original domain information from different modalities, serving as self-supervised information to refine the binary codes. Finally, CDAH skillfully uses a scaled tanh function to adaptively learn the binary codes, which will gradually converge to the original tricky binary coding problem. We conduct comprehensive experiments on four popular datasets, and the experimental results demonstrate the superiority of our model against the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Person re-identification (re-ID) has made substantial progress in recent years; however, it is still challenging to search for the target person in a short time. Re-ID with deep hashing is a shortcut for that but, limited by the expression of binary code, the performance of the hashing method is not satisfactory. Besides, to further speed up retrieval, researchers tend to reduce the number of feature bits, which will cause more performance degradation. In this paper, we design the attribute-based fast retrieval (AFR), which leverages the attribute prediction of the model trained in a binary classification manner tailor-made for hashing. The attribute information is also used to refine the global feature representation by an attribute-guided attention block (AAB). Then, to fully exploit deep feature to generate the hash codes, we propose a binary code learning method, named self-distilling smooth relaxation (SSR). In this method, a simple yet effective regularization is presented to distill the quantized knowledge in the model itself, thus mitigating the lack of semantic guidance in the traditional non-linear relaxations. We manually label attributes for each person in dataset CUHK03 and evaluate our method on four authoritative public benchmarks (Market-1501, Market-1501+500K, CUHK03, and DukeMTMC-reID). The experimental results indicate that with the SSR and AAB, we surpass all the state-of-the-art hashing methods. And compared with reducing the feature bits, the AFR strategy is more effective to save search time. [ABSTRACT FROM AUTHOR]

• A novel variant of softmax loss based on hard mining is proposed to FGIR. • The proposed loss can separate the closest subspecies of fine-grained image. • Extensive experiments on several classical benchmarks demenstrate its effectiveness. Although there is abundant of investigations on fine-grained image retrieval, it is still an extremely challenging task in the field of computer vision, due to the character of small diversity in inter-class but large diversity within intra-class. To handle this task, loss functions are critical to the performance of a deep convolutional neural network in extracting the discriminative feature of the fine-grained image for retrieval. Recent studies showed that the global structure loss functions help to extract more discriminative features. In this paper, we introduce a novel global structure loss function, named Hard Decorrelated Centralized Loss, for further improving the representation for fine-grained image retrieval. The proposed loss is available in extracting the discriminative feature for dividing the most similar categories. In our experiments, we employ the proposed loss to train the convolutional neural network, which shows state-of-the-art performances on six classical fine-grained image retrieval benchmarks, e.g. CUB-200-2011 and Stanford Cars. [ABSTRACT FROM AUTHOR]

Designers usually retrieve and review existing relevant product images to draw inspiration when designing a new proposal. Obtaining reasonable retrieval results of a given target is a challenging task due to the large-scale, low-correlation searching process in the product image field. In this paper, an end-to-end skip-connected network is introduced to improve the accuracy of retrieval by directly addressing the above problems. In more detail, skip-connections are employed to combine the multi-resolution feature maps produced by a pre-trained backbone network, which provides important improvements to address the presence of large scale variance problem. And the second-order information of the multi-resolution feature maps is obtained by using gram matrix operation, which allows the network to exploit more representative features and their correlations. Experimental results indicate that the proposed method achieves better retrieval performance when compared to current state-of-the-art techniques on there normal benchmark datasets as well as the product image data. [ABSTRACT FROM AUTHOR]

In this paper, a novel approach which can map high-dimensional, real-valued data into low-dimensional, binary vectors is proposed to achieve fast approximate nearest neighbor (ANN) search. In our paper, the binary codes are required to preserve the relative similarity, which makes the Hamming distances of data pairs approximate their Euclidean distances in ANN search. Under such constraint, the distribution adaptive binary labels are obtained through a lookup-based mechanism. The perpendicular bisector planes located between two kinds of data whose binary labels are different on only one specific bit are considered as weak hash functions. As just two kinds of data are taken into consideration during generation of the weak hash functions, the final strong hash functions are formed by combining the weak ones through boosting scheme to map all kinds of data into binary codes effectively. Experimental results show that our algorithm can encode the out of samples efficiently, and the performances of our method are superior to many state-of-the-art methods. [ABSTRACT FROM AUTHOR]

With the ever-increasing growth of massive high-dimensional data, deep learning to hash technology has been widely used for approximate nearest neighbor search on large-scale datasets, due to its remarkable efficiency and retrieval performance. In this paper, we propose a novel supervised deep hashing method, named Multi-granularity Feature Learning Hashing (MFLH), to learn compact binary descriptors. Specifically, we design an end-to-end trainable network to jointly learn feature representations and hash codes, in which a global stream and a local stream are responsible for learning feature representations with different granularities, and a hashing stream is devoted to encoding multi-granularity features into binary codes. In the local stream, a Cyclic Shift Mechanism (CSM) strategy is developed to assist mining more discriminative local fine information. In the meantime, an approximate sign activation function, which can be used for continuous optimization, is introduced to reduce quantization error. Furthermore, an improved variant of the triplet loss is presented to enhance the representation of pair-wise similarity for hash codes. Extensive experiments demonstrate that our proposed method significantly outperforms state-of-the-art hashing methods on the benchmark datasets, thereby verifying the effectiveness of our approach. Source code is provided for reproducibility. [ABSTRACT FROM AUTHOR]

Human identification is an important task that can help protect information security. Building deep learning models for human identification from Electrocardiogram (ECG) data is one of the highly promising technique. It has several unique advantages such as liveness detection, insensitive, easy to collect, higher security and so on. However, existing classifier-based methods only support closed-set identification, while existing matching-based methods are limited to high computational complexity. Besides, almost all methods only consider one-shot identification, which might be affected by occasional noise. In this paper, we propose CardioID to solve the above problems. CardioID learns binary codes from continuous ECG data which can identify faster than existing methods. It also supports identifying new person without the need to reconstructed or re-train the model. Besides, it can theoretically guarantee the recognition accuracy by introducing statistical hypothesis testing for making an identification decision. Experiments on real world ECG data show that CardioID can achieve 9.84% higher identification accuracy while saving 30.90% of running time compared with each of the second best baselines. [ABSTRACT FROM AUTHOR]

• Proposed a deep asymmetric hashing via unequal-dimensional embeddings. • Proposed a well-designed alternative optimization to solve the discrete problem. • Evaluated the effectiveness of the proposed method via three famous datasets. In recent years, Hashing has become a popular technique used to support large-scale image retrieval, due to its significantly reduced storage, high search speed and capability of mapping high dimensional original features into compact similarity-preserving binary codes. Although effectiveness achieved, most existing hashing methods are still some limitations, including: (1) Many supervised hashing methods only transform the label information into pairwise similarities to guide the hash code learning process, which will lead to the loss of rich semantic information between image pairs. (2) Some pioneer hashing methods use a relaxation-based strategy to solve discrete problems, resulting in a large quantization error. (3) Some supervised hashing methods handle the hashing learning procedure based on an asymmetric learning manner, and although this partially solves the problems of low efficiency and accuracy of symmetric learning strategy, they are all based on the embeddings of equal dimension, which leads to the reduction in the models representation ability and an increase in potential noise. To overcome the above limitations, in this paper, we propose a novel yet simple but effective hashing method, named Scalable Deep Asymmetric Hashing (SDAH). Specifically, SDAH is an end-to-end deep hashing method based on a fast iterative optimization strategy, which utilizes two real-valued embeddings of unequal dimensions, i.e., real-valued embeddings of images and labels, to flexibly perform asymmetric similarity computation. It can circumvent the use of the large pairwise similarity matrix by introducing an intermediate label matrix term which results in a remarkable reduction in the memory space cost. By doing this, the learned hash codes are more semantically informative for image retrieval tasks. Experimental results on several benchmark datasets highlight the superiority of SDAH in comparison with many state-of-the-art hashing methods. [ABSTRACT FROM AUTHOR]

For multi-label image retrieval based on deep hashing, the ultimate challenge is to map from the original image to binary space while preserving high-level semantic similarity. Recently, many supervised deep hashing approaches for multi-label image retrieval have been proposed to generate high-quality binary codes. However, most such methods are only introduced to learn simple similarity based on these image characteristics, therein ignoring complex multilevel semantic similarity with fine-grained features. In this paper, we propose a framework named deep hashing with fine-grained feature learning (DH-FFL) to preserve complex multilevel semantic similarity between multi-label image pairs. In this proposed model, compact bilinear pooling convolutional neural networks (CNNs) with normalization are adopted to extract fine-grained feature descriptors. In addition, a novel multilevel contrastive loss is designed to preserve multilevel semantic similarity by introducing a zero-loss parameter. Moreover, a multi-label classification loss is used to maintain the unique semantic structure of each image and maximize the distinguishing ability of binary codes. Comprehensive experiments on three benchmark datasets show that the proposed DH-FFL achieves promising performance compared with other state-of-the-art multi-label image retrieval applications. [ABSTRACT FROM AUTHOR]

Unsupervised cross-modal hashing learns hash codes without dependence on the semantic labels. It has the desirable advantage of well scalability and thus can effectively support large-scale cross-media retrieval. However, how to directly learn discriminative discrete hash codes under the unsupervised learning paradigm is still an open challenging problem. In this paper, we aim to attack this problem by proposing an Unsupervised Deep Cross-modal Hashing with Virtual Label Regression (UDCH-VLR) method. We propose a novel unified learning framework to jointly perform deep hash function training, virtual label learning and regression. Specifically, we learn unified hash codes via collaborative matrix factorization on the multi-modal deep representations to preserve the multi-modal shared semantics. Further, we incorporate the virtual label learning into the objective functions and simultaneously regress the learned virtual labels to the hash codes. Finally, instead of simply exploiting the existing shallow features and relaxing the binary constraints, we devise an alternative optimization strategy to directly update the deep hash functions and discrete binary codes. Under such circumstance, the discriminative capability of hash codes can be progressively enhanced with iterative learning. Extensive experiments on three publicly available cross-media retrieval datasets demonstrate that our approach outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Thanks to its storage and computation efficiency, hashing as a kind of nearest neighbor search method can facilitate massive data processing in recent vision and learning studies. Particularly, deep supervised hashing methods have significantly improved the retrieval performance compared with non-deep supervised hashing methods. However, most existing deep supervised hashing methods approximate the similarity between two images with the hamming distance between the outputs of the same hash function, i.e., the symmetric strategy. Consequently, it is typically time-consuming to train these symmetric hashing methods, and these methods can hardly take full advantage of the supervised information in the large-scale database. In this paper, we propose a novel discriminative deep m etric learning approach for a symmetric d iscrete h ashing (ADMH) approach for supervised hashing learning. ADMH integrates an asymmetric strategy with a deep metric learning method to learn the hash function for the query images and the discrete hash codes for database images directly. More specifically, we train a deep neural network to extract the features of the query images. Subsequently, we introduce a metric learning scheme for learning the discrete discriminative hash codes of the database images directly. Finally, the feature learning procedure for generating the hash codes of the query images and the discrete coding procedure for generating the hash codes of the database images are integrated into an end-to-end learning framework. Extensive experiments on various benchmark datasets show that the proposed asymmetric deep hashing method outperforms the existing hashing methods. [ABSTRACT FROM AUTHOR]

An urgent problem in the field of deep learning is the optimization of model construction, which frequently hinders its performance and often needs to be designed by experts. Optimizing the hyper-parameters remains a substantial obstacle in designing deep learning models, such as CNNs, in practice. In this paper, we propose an automatical optimization framework using binary coding system and GPSO with gradient penalties to select the structure. Such swarm intelligence optimization approaches have been used but not extensively exploited, and the existing work focuses on models with a fixed depth of networks. We design an experiment to arouse three types of emotion states for each subject, and simultaneously collect EEG signals corresponding to each emotion category. The GPSO-based method efficiently explores the solution space, allowing CNNs to obtain competitive classification performance over the dataset. Results indicate that our method based on the GPSO-optimized CNN model enables us to achieve a prominent classification accuracy, and the proposed method provides an effective automatic optimization framework for CNNs of the emotion recognition task with an uncertain depth of networks. [ABSTRACT FROM AUTHOR]

Several visual features have been developed for content-based image retrieval in the last decades, including global, local and deep learning-based approaches. However, despite the huge advances in features development and mid-level representations, a single visual descriptor is often insufficient to achieve effective retrieval results in several scenarios. Mainly due to the diverse aspects involved in human visual perception, the combination of different features has been establishing as a relevant trend in image retrieval. An intrinsic difficulty consists in the task of selecting the features to combine, which is often supported by supervised learning approaches. Therefore, in the absence of labeled data, selecting features in an unsupervised way is a very challenging, although essential task. In this paper, an unsupervised framework is proposed to select and fuse visual features in order to improve the effectiveness of image retrieval tasks. The framework estimates the effectiveness and correlation among features through a rank-based analysis and uses a list of ranker pairs to determine the selected features combinations. High-effective retrieval results were achieved through a comprehensive experimental evaluation conducted on 5 public datasets, involving 41 different features and comparison with other methods. Relative gains up to +55% were obtained in relation to the highest effective isolated feature. [ABSTRACT FROM AUTHOR]

Unsupervised deep hash functions are complicated due to the challenges of learning discriminative clusters and the absence of similarity-sensitive objectives. Existing approaches predominately handle these challenges independently, while neglecting the fact that cluster centroids and similarity-sensitive binary codes are correlated and can be learned simultaneously. In this paper, we propose a novel end-to-end deep framework for image retrieval, namely Clustering-driven Unsupervised Deep Hashing (CUDH), to recursively learn discriminative clusters by soft clustering model and produce binary code with high similarity responds. We employ the aggregated clusters as an auxiliary distribution to generate hashing codes. With imposing binary constraints loss and reconstruction loss of auto-encoder, our CUDH can be jointly optimized by standard stochastic gradient descent (SGD). Comprehensive experiments on three popular datasets are conducted and the results show that our CUDH can outperform the state-of-the-art methods by large margins. [ABSTRACT FROM AUTHOR]

This paper aims to develop a new local binary learning feature for finger vein images called personalized binary code (PBC). Existing local learning features of the image, such as compact binary face descriptor (CBFD) and discriminant binary descriptor (DBD), ignore the structure of the binary feature corresponding to each class when learning. Inspired by the bit-consistency phenomenon of local binary pattern (LBP), we combine the constraints of within-class sparse constraints with linear discriminant analysis (LDA) criteria in the learning process to explore the bit-consistent structure of learned binary features. Specifically, a multi-directional pixel difference vector is first calculated at each pixel of the training set of finger vein images, and an objective function is designed to learn a discriminative map that projects these multi-directional pixel difference vectors into binary features. Our objective function consists of several constraints: Minimizing quantization error, minimizing within-class variance of learned binary features, and maximizing the between-class variance of learned binary features, and minimizing the L 2,1 norm of the learned features for each class. Next, the algorithm based on alternating direction method (ADM) is designed to solve the optimization problem. Finally, a histogram representation of each finger vein image is constructed by the bag-of-words model. In our experiments, the proposed PBC shows that the leading performance on two public databases in comparison with existing finger vein recognition methods. [ABSTRACT FROM AUTHOR]

Hashing methods are used to perform the approximate nearest neighbor search due to the low storage for binary codes and the fast computation of Hamming distance. However, in most of the hashing methods, the learning process of hash functions has high cost in both time and storage. To overcome this issue, in this paper, a fast online unsupervised hashing method based on data sampling is proposed to learn the hypersphere-based hash functions from the streaming data. By maintaining a small-size data sample to efficiently preserve the properties of the streaming data, the hypersphere-based hash functions are learnt in an online fashion from the data sample and we can justify the hash functions by proving their theoretic properties. To further improve the search accuracy of our method, a new dimensionality reduction algorithm is proposed to learn the projection matrix from the data sample to construct a low-dimensional space. Then, the data sample is projected into the low-dimensional space, and our method can learn the hash functions online from the small-size projected data sample with low computational complexity and storage space. The experiments show that our method has a better search accuracy than other online hashing methods and runs faster in learning the hash functions. [ABSTRACT FROM AUTHOR]

Content-based Medical Image Retrieval (CBMIR) presents promising results in computer-aided diagnosis which provides the clinicians with interpretative guidance based on the visual similarity. In this paper, we focus on the tasks of histopathological image retrieval for breast cancer diagnosis. Instead of the holistic-based and cell-based methods proposed in previous works, we propose a multi-magnification correlation hashing (MMCH) framework that learns the discriminative binary codes for histopathological images. The patch-link graph is constructed with both low and high magnification images where the semantic links are propagated from the pre-labeled patches to unlabeled data. We then learn the binary hashing codes based on the global labels from the low-magnification images with structure-regularization. The experiments on BreakHis dataset and PLOSONE dataset demonstrate that MMCH outperforms the previous hashing methods on histopathological image retrieval. [ABSTRACT FROM AUTHOR]

In recent years, the binary feature descriptor has achieved great success in face recognition (FR) field, such as local binary pattern (LBP). It is well known that the high-dimensional feature representations can contain more discriminative information, therefore, it is natural for us to construct the high-dimensional binary feature for FR task. However, the high-dimensional representations would lead to high computational cost and overfitting. Therefore, an effective sparsity regularizer is necessary. In this paper, we introduce the sparsity constraint into the objective function of general binary codes learning framework, so that the problem of high computational cost and overfitting can be somehow solved. There are three main requirements in our objective function: First, we require that the high-dimensional binary codes have the minimized quantization loss compared with centered original data. Second, we require the projection matrices are sparse, so that the projection process would not take lots of computational resource even faced with high-dimensional original data. Third, for a mapping (hashing) function, the bit-independence and bit-balance are two excellent properties for generating discriminative binary codes. We also empirically show that the high-dimensional binary codes can obtain more discriminative ability by pooling process with an unsupervised clustering method. Therefore, a discriminative and low-cost Sparse Projection Matrix Binary Descriptors (SPMBD) is learned by the data-driven way. Extensive experimental results on four public datasets show that our SPMBD descriptor outperforms other existing face recognition algorithms and demonstrate the effectiveness and robustness of the proposed methods. [ABSTRACT FROM AUTHOR]

Recent years have witnessed the promising capacity of hashing techniques in tackling nearest neighbor search because of the high efficiency in storage and retrieval. Data-independent approaches (e.g., Locality Sensitive Hashing) normally construct hash functions using random projections, which neglect intrinsic data properties. To compensate this drawback, learning-based approaches propose to explore local data structure and/or supervised information for boosting hashing performance. However, due to the construction of Laplacian matrix, existing methods usually suffer from the unaffordable training cost. In this paper, we propose a novel supervised hashing scheme, which has the merits of (1) exploring the inherent neighborhoods of samples; (2) significantly saving training cost confronted with massive training data by employing approximate anchor graph; as well as (3) preserving semantic similarity by leveraging pair-wise supervised knowledge. Besides, we integrate discrete constraint to significantly eliminate accumulated errors in learning reliable hash codes and hash functions. We devise an alternative algorithm to efficiently solve the optimization problem. Extensive experiments on various image datasets demonstrate that our proposed method is superior to the state-of-the-arts. [ABSTRACT FROM AUTHOR]

Band selection is one of the most important topics in hyperspectral image classification for irrelevant band information and the high correlation between the adjacent bands. The main concern is to obtain the compact and effective bands to classify the image with the least impact for the classification accuracy. In general, band selection could be seen as a combinatorial optimization problem through defining an objective function based on the number of bands and classification accuracy. Therefore, in the paper, a novel band selection method based on a chaotic binary coded gravitational search algorithm (CBGSA) is proposed to reduce the dimensionality of airborne hyperspectral images. The proposed method is also compared with that of genetic algorithm (GA), binary coded particle swarm optimization (BPSO) algorithm, binary coded differential evolution (BDE) algorithm and binary coded cuckoo search (BCS) algorithm on some airborne hyperspectral images; furthermore, it is also compared with some other existing techniques such as Relief-F algorithm, minimum Redundancy Maximum Relevance (mRMR) criterion, and the optimum index (OI) criterion for a comprehensive comparison. Experimental results display that the proposed method is robust, adaptive and might be applied for practical work of airborne hyperspectral image classification. [ABSTRACT FROM AUTHOR]

Recent studies have shown that image representations built upon deep convolutional layers in Convolutional Neural Networks (CNNs) have strong discriminative characteristics. In this paper, we present a novel and effective method to create compact binary codes (deep binary codes) based on deep convolutional features for image retrieval. Deep binary codes are generated by comparing the response from each feature map and the average response across all the feature maps on the deep convolutional layers. Additionally, a spatial cross-summing strategy is proposed to directly generate bit-scalable binary codes. As the deep binary codes on different deep layers can be obtained by passing the image through the CNN and each of them makes a different contribution to the search accuracy, we then present a dynamic, on-the-fly late fusion approach where the top N high quality search scores from deep binary codes are automatically determined online and fused to further enhance the retrieval precision. Two strengths of the proposed methods are that the generation of deep binary codes is based on a generic model, which does not require additional training for new image domains, and that the dynamic late fusion scheme is query adaptive. Extensive experimental results on well known benchmarks show that the performance of deep binary codes are competitive with state-of-the-art approaches for large scale image retrieval. Moreover, it is shown that the dynamic late fusion scheme significantly enhances the search accuracy. [ABSTRACT FROM AUTHOR]

There has been increasing interest in learning compact binary codes for large-scale image data representation and retrieval. In most existing hashing-based methods, high-dimensional vectors are hashed into Hamming space, and the similarity between two vectors is approximated by the Hamming distance between their binary codes. Although hashing-based binary codes generation methods were widely used, Product Quantization (PQ) has been shown to be more accurate than various hashing-based methods, largely due to its lower quantization distortions and more precise distance computation. However, it is still a challenging problem to generalize PQ to accommodate arbitrary kernels. In this paper, we demonstrate how to employ arbitrary kernel functions in a PQ scheme. First, we propose a Kernelized PQ (KPQ) method based on composite kernels, which serves as a basic framework by making the decomposition of implicit feature space possible. Furthermore, we propose a Kernelized Optimized PQ (KOPQ) method to generalize Optimized Product Quantization (OPQ) to an arbitrary implicit feature space. Finally, we propose a Supervised KPQ (SKPQ) to improve the performance of semantic neighbor search. Both methods are variations of KPQ with the incorporation of their corresponding core techniques, KPCA and KCCA respectively, to the basic KPQ framework. Experiments involving three notable datasets show that KPQ, KOPQ and SKPQ can outperform the state-of-the-art methods for a similarity search in feature space or semantic search. [ABSTRACT FROM AUTHOR]

Neuroscientists formulate very different hypotheses about the nature of neural coding. At one extreme, it has been argued that neurons encode information through relatively slow changes in the arrival rates of individual spikes (rate codes) and that the irregularity in the spike trains reflects the noise in the system. At the other extreme, this irregularity is the code itself (temporal codes) so that the precise timing of every spike carries additional information about the input. It is well known that in the estimation of Shannon Information Transmission Rate, the patterns and temporal structures are taken into account, while the “rate code” is already determined by the firing rate, i.e. by the spike frequency. In this paper we compare these two types of codes for binary Information Sources, which model encoded spike trains. Assuming that the information transmitted by a neuron is governed by an uncorrelated stochastic process or by a process with a memory, we compare the Information Transmission Rates carried by such spike trains with their firing rates. Here we show that a crucial role in the relation between information transmission and firing rates is played by a factor that we call the “jumping” parameter. This parameter corresponds to the probability of transitions from the no-spike-state to the spike-state and vice versa . For low jumping parameter values, the quotient of information and firing rates is a monotonically decreasing function of the firing rate, and there therefore a straightforward, one-to-one, relation between temporal and rate codes. However, it turns out that for large enough values of the jumping parameter this quotient is a non-monotonic function of the firing rate and it exhibits a global maximum, so that in this case there is an optimal firing rate. Moreover, there is no one-to-one relation between information and firing rates, so the temporal and rate codes differ qualitatively. This leads to the observation that the behavior of the quotient of information and firing rates for a large jumping parameter value is especially important in the context of bursting phenomena. [ABSTRACT FROM AUTHOR]

Binary feature descriptors have been widely used in computer vision field due to their excellent discriminative power and strong robustness, and local binary patterns (LBP) and its variations have proven that they are effective face descriptors. However, the forms of such binary feature descriptors are predefined in the hand-crafted way, which requires strong domain knowledge to design them. In this paper, we propose a simple and efficient iterative quantization binary codes (IQBC) feature learning method to learn a discriminative binary face descriptor in the data-driven way. Firstly, similar to traditional LBP method, we extract patch-wise pixel difference vectors (PDVs) by computing and concatenating the difference between center patch and its neighboring patches. Then, inspired by multi-class spectral clustering and the orthogonal Procrustes problem, which both are widely used in image retrieval field, we learn an optimized rotation to minimize the quantization error of mapping data to the vertices of a zero-centered binary hypercube by using iterative quantization scheme. In other words, we learn a feature mapping to project these pixel difference vectors into low-dimensional binary vectors. And our IQBC can be used with unsupervised data embedding method such as principle component analysis (PCA) and supervised data embedding method such as canonical correlation analysis (CCA), namely IQBC-PCA and IQBC-CCA. Lastly, we cluster and pool these projected binary codes into a histogram-based feature that describes the co-occurrence of binary codes. And we consider the histogram-based feature as our final feature representation for each face image. We investigate the performance of our IQBC-PCA and IQBC-CCA on FERET, CAS-PEAL-R1, LFW and PaSC databases. Extensive experimental results demonstrate that our IQBC descriptor outperforms other state-of-the-art face descriptors. [ABSTRACT FROM AUTHOR]

Nowadays the amount of multimedia data such as images and text is growing exponentially on social websites, arousing the demand of effective and efficient cross-modal retrieval. The cross-modal hashing based methods have attracted considerable attention recently as they can learn efficient binary codes for heterogeneous data, which enables large-scale similarity search. Generally, to effectively construct the cross-correlation between different modalities, these methods try to find a joint abstraction space where the heterogeneous data can be projected. Then a quantization rule is applied to convert the abstraction representation to binary codes. However, these methods may not effectively bridge the semantic gap through the latent abstraction space because they fail to capture latent information between heterogeneous data. In addition, most of these methods apply the simplest quantization scheme (i.e. sign function) which may cause information loss of the abstraction representation and result in inferior binary codes. To address these challenges, in this paper, we present a novel cross-modal hashing based method that generates unified binary codes combining different modalities. Specifically, we first extract semantic features from the modalities of images and text to capture latent information. Then these semantic features are projected to a joint abstraction space. Finally, the abstraction space is rotated to produce better unified binary codes with much less quantization loss, while preserving the locality structure of projected data. We integrate the binary code learning procedures above to develop an iterative algorithm for optimal solutions. Moreover, we further exploit the useful class label information to reduce the semantic gap between different modalities to benefit the binary code learning. Extensive experiments on four multimedia datasets show that the proposed binary coding schemes outperform several other state-of-the-art methods under cross-modal scenarios. [ABSTRACT FROM AUTHOR]

Binary code embedding, which aims to generate compact and discriminative binary codes from local image features, can remarkably improve the image search performance by compensating the quantization error in Bag-of-Words (BoW) model. However, the relationship between local features and their neighbors are often ignored by existing embedding schemes, while such information of spatial distribution can greatly improve the discriminative ability of binary codes. Toward this end, this paper proposes two light-weight schemes for binary code embedding that take the spatial distribution of local features into account. These two schemes, including the Content Similarity Embedding (CSE) and Scale Similarity Embedding (SSE), are highly flexible in balancing the computational cost as well as the image search performance. Experimental results on several public benchmarks show that, with the proposed two embedding schemes, image search achieves comparable performance with state-of-the-arts with much lower computational cost and memory usage. [ABSTRACT FROM AUTHOR]

Hashing methods map high-dimensional data onto compact binary codes for efficient retrieval. These methods focus on preserving the data similarity in Hamming distance between the mapped hash codes. In this paper we propose a novel hashing method motivated by maximizing the probability of data with the same hash code being true neighbors, under the constraint of code compactness. This method is data-dependent and generates quite compact hash codes. The key idea is to use a collection of tree-structured hyperplanes to satisfy the compactness constraint, as well as to maximize the lower bound of the objective function. We compare our method with some widely used hashing methods on real datasets of different sizes. The experimental results illustrate the superior performance of our method. The performance of this method is further effectively improved by a multi-table extension. [ABSTRACT FROM AUTHOR]

Abstract: Due to explosive growth of visual content on the web, there is an emerging need of fast similarity search to efficiently exploit such enormous web contents from very large databases. Recently, hashing has become very popular for efficient nearest neighbor search in large scale applications. However, many traditional hashing methods learn the binary codes in a single shot or only employ a single hash table, thus they usually cannot achieve both high precision and recall simultaneously. In this paper, we propose a novel dual complementary hashing (DCH) approach to learn the codes with multiple hash tables. In our method, not only the projection for each bit inside a hash table has the property of error-correcting but also the different hash tables complement each other. Therefore, the binary codes learned by our approach are more powerful for fast similarity search. Extensive experiments on publicly available datasets demonstrate the effectiveness of our approach. [Copyright &y& Elsevier]

Abstract: This paper introduces a two-stage strategy for multi-class classification problems. The proposed technique is an advancement of tradition binary decomposition method. In the first stage, the classifiers are trained for each class versus the remaining classes. A modified fitness value is used to select good discriminators for the imbalanced data. In the second stage, the classifiers are integrated and treated as a single chromosome that can classify any of the classes from the dataset. A population of such classifier-chromosomes is created from good classifiers (for individual classes) of the first phase. This population is evolved further, with a fitness that combines accuracy and conflicts. The proposed method encourages the classifier combination with good discrimination among all classes and less conflicts. The two-stage learning has been tested on several benchmark datasets and results are found encouraging. [Copyright &y& Elsevier]