Search

In this paper, we propose intra- and inter-feature orthogonal fusion (IIOF) of local and global features obtained from MS-SincResNet or MS-SSincResNet (a variant of MS-SincResNet) for music emotion recognition (MER). Given a raw waveform of music signal, MS-SincResNet/MS-SSincResNet is first used to learn several 2D representations having different receptive fields and obtain embeddings with time-frequency information from different layers. Then, local and global features are extracted from these embeddings. IIOF consisting of intra-feature OF and inter-feature OF is further employed to integrate both local and global features to obtain a discriminative descriptor for MER. The intra-feature OF is used to enhance the diversity of the global feature, and the inter-feature OF is utilized to reduce redundancies and produce complementary information between local and global features. The experimental results have demonstrated that the representation discriminability can be enhanced by IIOF considering the feature orthogonality. Furthermore, extensive experimental results have shown that the proposed method outperforms other state-of-the-art methods in terms of regression and classification tasks on the well-known MER datasets, including the DEAM dataset and the PMEmo dataset. The codes are available at https://github.com/PeiChunChang/MS-SSincResNet_with_IIOF. • We propose IIOF approach to effectively integrating local and global features. • IIOF is combined with MS-SincResNet and MS-SSincResNet for music emotion recognition. • The proposed method outperforms other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Hashing algorithm has been widely used for efficient approximate nearest neighbor (ANN) search. Learning optimal hashing functions has been given focus and it is still a challenge. This paper aims to effectively and efficiently generate relative similarity preserving binary codes. Most existing hashing methods try to preserve the locality similarity by preserving direct distance similarity, while ignoring the relative similarity which advantages in ANN search. In this paper, this issue is solved by proposing the relative error which emphasizes that the ordinal relations in Hamming space and Euclidean space should be consistent with each other. We learn hashing projection functions via two steps. The first step adopts the lookup-based mechanism to find the optimal binary codes of training data, which can preserve the relative similarity and simultaneously adapt to data distribution. The binary codes in the first step are considered as supervision information in the second step. The objective of the second step is to learn hashing projection functions which can efficiently regenerate the binary codes in the first step. Aim to be in accordance with the property of data distribution, the hyper internal tangent planes of two specified spheres are chosen as hashing projection functions. Assisted by these projection functions, the time complexity of encoding process is greatly reduced. Experimental results on four public data sets demonstrate that our method outperforms many other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Highlights • A weighted similarity matrix minimizing the disparity between the similarities in the original space and the similarities in the hashing space. • Hashing function obtained by multilayer neural network and optimized by an efficient way. • Quantization errors between the input feature and the learned compact binary codes reduced by the proposal. Abstract Learning based hashing technologies have been widely adopted in multimedia retrieval as they could afford efficient storage and extract semantic information for high-dimensional data. However, the major difficulty of learning based hashing is the discrete constraint imposed on the required hashing codes, which makes the optimization generally to be NP-hard. In this paper, a novel supervised learning based discrete hashing (SLDH) approach is proposed to learn compact binary codes under the deep learning framework for image retrieval. We adopt multilayer network to convert the original features into binary codes, while it should exploit the semantic relevance of manual labels and keep the semantic similarity. For this purpose, we propose the objective function to obtain the binary codes including: 1) making full use of manual labels to get implicit semantic information; 2) using the weighted similarity matrix to keep the semantic similarity; 3) relaxing the discrete constraint to a normalized optimization problem; 4) adding the orthogonality constraint on binary codes to reduce the information redundancy. The objective function is optimized with the alternating direction method and modified alternating direction of multipliers(ADMM) algorithm with efficient iteration. Experiments are conducted on three databases and the results demonstrate the superiority to several state-of-the-art hashing based image retrieval methods. [ABSTRACT FROM AUTHOR]

• Propose a deep video code (DVC) framework to represent video face as binary codes. • Propose two video modeling schemes to effectively aggregate information from video frames. • A novel bounded triplet loss is elaborately designed for discriminative hash learning. • Demonstrate superior performance of DVC for both video to video and image to video retrieval tasks on three challenging datasets. In this paper, we address one specific video retrieval problem in terms of human face. Given one query in forms of either a frame or a sequence from a person, we search the database and return the most relevant face videos, i.e., ones have the same class label with the query. Such problem is very challenging due to the large intra-class variations and the high request on the efficiency of video representations in terms of both time and space. To handle such challenges, this paper proposes a novel Deep Video Code (DVC) method which encodes video faces into compact binary codes. Specifically, we devise an end-to-end convolutional neural network (CNN) framework that takes face videos as training inputs, models each of them as a unified representation by temporal feature pooling operation, and finally projects the high-dimensional representations of both frames and videos into Hamming space to generate binary codes. In such Hamming space, distance of dissimilar pairs is larger than that of similar pairs by a margin. To this end, a novel bounded triplet hashing loss is elaborately designed, which takes all dissimilar pairs into consideration for each anchor point in a mini-batch, and the optimization of the loss function is smoother and more stable. Extensive experiments on challenging video face databases and general image/video datasets with comparison to the state-of-the-arts verify the effectiveness of our method in different kinds of retrieval scenarios. [ABSTRACT FROM AUTHOR]

In this paper, we present a new cross-modal discrete hashing (CMDH) approach to learn compact binary codes for cross-modal multimedia search. Unlike most existing cross-modal hashing methods which usually relax the optimization objective function to obtain hash codes, we develop a discrete optimization framework to jointly learn binary codes and a series of hash functions for each modality, so that the performance drop due to the inferior optimization techniques can be avoided. Specifically, we present two cross-modal hashing algorithms called CMDH-linear and CMDH-kernel under the proposed framework, which performs linear and non-linear mappings to learn binary codes, respectively. Different from existing cross-modal hashing methods which maximize the corrections of hash codes from different modalities, our CMDH learns a set of shared binary codes for samples captured from different modalities, so that the modality gap can be effectively removed in cross-modal multimedia retrieval. To further improve the flexibility of our approach for different scenarios, we extend CMDH to unsupervised CMDH (unCMDH) and discrete multi-modal hashing (MMDH), which learns hash codes for training data without label information and with multi-modal labelled data. Experimental results on three benchmark datasets clearly show that our methods achieve competitive results with the state-of-the-arts. [ABSTRACT FROM AUTHOR]

In the context of large scale similarity search, one promising technique is to encode high dimensional data as compact binary codes to take advantage of the speed and storage efficiencies. Many existing hashing approaches achieve similarity preservation in the Hamming space by preserving similarity relationship between data points. However, most of these methods only consider the relationship between points, which can not capture the data structure comprehensively. In this paper, we propose a reconstruction-based supervised hashing (RSH) method to learn compact binary codes with holistic structure preservation. The proposed method characterizes the similarity structure by the relationship between each data point and the structure generated by the remaining points. The learning objective is set to simultaneously minimize the distance between each point and the structure with the same class label and maximize the distance between each point and the structure with different class labels. In cross-modal retrieval, we propose a reconstruction-based hashing method by distilling the correlation structure in the common latent hamming space. The correlation structure characterizes the semantic correlation by the relationship between data points and structures in the common hamming space. Minimizing the reconstruction error of each single-modal latent model makes hidden layer outputs representative for the input of each modality. Experimental results in both single-modal and cross-modal datasets demonstrate the effectiveness of our methods when compared to several recently proposed approaches. [ABSTRACT FROM AUTHOR]

• Decoupling the code generation and hash learning is one way to continually hashing. • Label- hash code pairs are maintained to keep the encoding ability for seen concepts. • The worth-storing data is picked out by recording loss trajectory of each sample. • Gradient projection improves the probability of codes unchanged after model updated. Cross-modal hashing (CMH) has become widely used for large-scale multimedia retrieval. However, most current CMH methods focus on the closed retrieval scenario, not the real-world environments, i.e., complex and changing semantics. When data containing new class objects emerge, the current CMH has to retrain the model on all history training data, not the new data, to accommodate new semantics, but the never-stop upload of data on the Internet makes this impractical. In this paper, we devise a deep hashing method called Continual Cross-Modal Hashing with Gradient Aware Memory (CCMH-GAM) for learning binary codes of multi-label cross-modal data with increasing categories. CCMH-GAM is a two-step hashing architecture, one hashing network learns to hash the increasing semantics of data, i.e., label, into the semantic codes, and other modality-specific hashing networks learn to map data into the corresponding semantic codes. Specifically, to keep the encoding ability for old semantics, a regularization based on accumulating low-storage label-code pairs is designed for the former network. For the modality-specific networks, we propose a memory construction method via approximating the full episodic gradients of all data by some exemplars and derive its fast implementation with the upper bound of approximation error. Based on this memory, we propose a gradient projection method to theoretically improve the probability of old data's code being unchanged after updating the model. Extensive experiments on three datasets demonstrate that CCMH-GAM can continually learn hash functions and yield state-of-the-art retrieval performance. [ABSTRACT FROM AUTHOR]

• A novel binary multi-view clustering approach is proposed. • A global criterion directly provides the cluster assignments. • The pairwise importance of each sample and view is automatically determined. • Bidirectional FFT of deep features initialize the binary codes. • Performance is compared against 12 competing methods on 4 multi-view image datasets. Clustering is inherently a process of exploratory data analysis. It has attracted more attention recently because much real-world data consists of multiple representations or views. However, it becomes increasingly problematic when dealing with large and heterogeneous data. It is worth noting that several approaches have been developed to increase computational efficiency, although most of them have some drawbacks: (1) Most existing techniques consider equal or static weights to quantify importance across different views and samples, so common and complementary features cannot be used. (2) The clustering task is performed by arbitrary initialization without caring about the rich structure of the joint discrete representation, and thus poorly executed. In this paper, we propose a novel approach called "Auto-Weighted Binary Multi-View Clustering Via Deep Initialization" for large-scale multi-view clustering based on two main scenarios. First, we consider the distinction between different views based on the importance of samples, and therefore apply a dynamic learning strategy for the automatic weighting of views and samples. Second, in the context of initializing binary clustering, we develop a new CNN feature and use a low-dimensional binary embedding by exploiting the efficient capabilities of Fourier mapping. Moreover, our approach simultaneously learns a joint discrete representation and performs direct clustering using a constrained binary matrix factorization; the optimization problem is perfectly solved in a unified learning model. Experimental results conducted on several challenging datasets demonstrate the effectiveness and superiority of the proposed approach over state-of-the-art methods in terms of accuracy, normalized mutual information, and purity. [ABSTRACT FROM AUTHOR]

• We leverage the background variations in a finger vein image as a new auxiliary feature. • We propose a new descriptor named Intensity Orientation Vector (IOV) for background cues. • We propose a new supervised binary-feature learning scheme to jointly learn compact binary codes from both vein traits and background cues. • Extensive experiments demonstrate the effectiveness of IOV and the binary-feature learning scheme. Recently, finger vein based biometric authentication has attracted considerable attention due to its high efficiency and high security. However, most existing finger vein representation methods focus on vein traits while ignoring background cues, although background cues also convey identity information specific to each individual. In this paper, we leverage background intensity variations in finger vein images as new features to enrich discriminative representation, and accordingly propose a new descriptor named Intensity Orientation Vector (IOV). IOV, scaleable to reflect characteristics of finger tissues, offers additional informative cues for finger vein representation. Furthermore, we propose a new learning scheme named Semantic Similarity Preserved Discrete Binary Feature Learning (SSP-DBFL) for finger vein recognition. Unlike the most bimodal binary feature representation methods, SSP-DBFL preserves high-level semantic similarity in a common Hamming space to exploit the consensus between vein traits and background cues. Specifically, given a finger vein image, we first extract the direction difference vectors (DDV) as the main vein traits and the IOV as the auxiliary background cues. Subsequently, we jointly learn projection functions from these two types of features in a supervised manner, converting the two features into discriminative binary codes with their semantic similarity preserved. Finally, the binary codes are pooled into histogram-based vectors for finger vein representation. Extensive experiments are conducted on five widely used finger vein databases and demonstrate the effectiveness of our proposed IOV and SSP-DBFL. [ABSTRACT FROM AUTHOR]

Finger vein recognition has drawn increasing attention from biometrics community due to its security and convenience. In this paper, a novel discriminative binary codes (DBC) learning method is proposed for finger vein recognition. First of all, subject relation graph is built to capture correlations among subjects. Based on the relation graph, binary templates are transformed to describe vein characteristics of subjects. To ensure that templates are discriminative and representative, graph transform is formulated into an optimization problem, in which the distance between templates from different subjects is maximized and templates provide maximum information about subjects. At last, supervised information for training instances is provided by the obtained binary templates, and SVMs are trained as the code learner for each bit. Compared with existing binary codes for finger vein recognition, DBC are more discriminative and shorter. In addition, they are generated with considering the relationships among subjects which may be useful to improve performance. Experimental results on PolyU database and MLA database demonstrate the effectiveness and efficiency of DBC for finger vein recognition and retrieval. [ABSTRACT FROM AUTHOR]

• A novel end-to-end deep rank hashing (DRH) network is proposed for optimal identification and rank hashing goals. The network is trained independently from the enrollee face image to avoid security leakage and retraining for newly enrolled users. • A pairwise margin-based angular loss, code balancing loss, and quantization error corrector designed for nonlinear subspace ranking hashing are proposed. Based on the combination of proposed loss functions, the DRH learns to generate compact, high discriminative, and consistent binary hash codes under open-set settings. • A novel cancellable identification scheme is proposed based on the one-time XOR cipher notion. • Comprehensive experiments and security analyses are carried out based on the five face datasets verification, open-set, and closed-set identification protocols. Cancellable biometrics (CB) is one of the major approaches for biometric template protection. However, almost all the prior arts are designed to work under verification (one-to-one matching). This paper proposes a deep learning-based cancellable biometric scheme for face identification (one-to-many matching). Our scheme comprises two key ingredients: a deep rank hashing (DRH) network and a cancellable identification scheme. The DRH network transforms a raw face image into discriminative yet compact face hash codes based upon the nonlinear subspace ranking notion. The network is designed to be trained for both identification and hashing goals with their respective rich identity-related and rank hashing relevant loss functions. A modified softmax function is utilized to alleviate the hashing quantization error, and a regularization term is designed to encourage hash code balance. The hash code is binarized, compressed, and secured with the randomized lookup table function. Unlike prior CB schemes that require two input factors for verification, the proposed scheme demands no additional input except face images during identification, yet the face template is replaceable whenever needed based upon a one-time XOR cipher notion. The proposed scheme is evaluated on five public unconstrained face datasets in terms of verification, closed-set and open-set identification performance accuracy, computation cost, template protection criteria, and security. [ABSTRACT FROM AUTHOR]

High dimensional dense features have been shown to be useful for face recognition, but result in high query time when searching a large-scale face database. Hence binary codes are often used to obtain fast query speeds as well as reduce storage requirements. However, binary codes for face features can become unstable and unpredictable due to face variations induced by pose, expression and illumination. This paper proposes a predictable hash code algorithm to map face samples in the original feature space to Hamming space. First, we discuss the ‘predictability’ of hash codes for face indexing. Second, we formulate the predictable hash coding problem as a non-convex combinatorial optimization problem, in which the distance between codes for samples from the same class is minimized while the distance between codes for samples from different classes is maximized. An Expectation Maximization method is introduced to iteratively find a sparse and predictable linear mapping. Lastly, a deep feature representation is learned to further enhance the predictability of binary codes. Experimental results on three commonly used face databases demonstrate the superiority of our predictable hash coding algorithm on large-scale problems. [ABSTRACT FROM AUTHOR]

In this paper, we present a new structured light 3D reconstruction method that can be applied to scenes in the presence of global illumination such as inter-reflection, subsurface scattering and severe projector defocus. The proposed method takes advantage of important properties embedded in the binary code patterns, and determines the minimum stripe width that eliminates the effects of subsurface scattering and projector defocus in the capturing stage. Moreover, errors caused by inter-reflection are detected, and corrected by an iterative approach. Our method can be applied to scenes where more than one global illumination effect appears in a scene point. The accuracy of our method is demonstrated by quantitative evaluation, and its robustness by qualitative evaluation when applying to real-world scenes with various characteristics, both of which are better than two of the currently known best methods. [ABSTRACT FROM AUTHOR]

• We are the first to explore the uncertainty of hashing bits during approximate optimization. Depending on the magnitude of uncertainty, the corresponding hashing bits and input images receive different attention. • We propose to explicitly model bit-level and image-level uncertainty, resorting to the output discrepancy between the hashing network and the momentum-updated network. • Extensive experiments on the CIFAR-10, the NUS-WIDE, the MS-COCO, and the largescale Clothing1M datasets demonstrate that our method significantly improves retrieval performance when compared with state-of-the-art methods. Combinatorial optimization (CO) has been a hot research topic because of its theoretic and practical importance. As a classic CO problem, deep hashing aims to find an optimal code for each data from finite discrete possibilities, while the discrete nature brings a big challenge to the optimization process. Previous methods usually mitigate this challenge by binary approximation, substituting binary codes for real-values via activation functions or regularizations. However, such approximation leads to uncertainty between real-values and binary ones, degrading retrieval performance. In this paper, we propose a novel Deep Momentum Uncertainty Hashing (DMUH). It explicitly estimates the uncertainty during training and leverages the uncertainty information to guide the approximation process. Specifically, we model bit-level uncertainty via measuring the discrepancy between the output of a hashing network and that of a momentum-updated network. The discrepancy of each bit indicates the uncertainty of the hashing network to the approximate output of that bit. Meanwhile, the mean discrepancy of all bits in a hashing code can be regarded as image-level uncertainty. It embodies the uncertainty of the hashing network to the corresponding input image. The hashing bit and image with higher uncertainty are paid more attention during optimization. To the best of our knowledge, this is the first work to study the uncertainty in hashing bits. Extensive experiments are conducted on four datasets to verify the superiority of our method, including CIFAR-10, NUS-WIDE, MS-COCO, and a million-scale dataset Clothing1M. Our method achieves the best performance on all of the datasets and surpasses existing state-of-the-art methods by a large margin. [ABSTRACT FROM AUTHOR]

l We propose a joint multi-view feature learning method for FKP recognition. l We form multi-view data vectors (MVDVs) to precisely describe a FKP image. l We jointly learn the compact hash codes of MVDVs in an unsupervised manner. l The JLCMHC method achieves promising performance for few-shot FKP recognition. As a relatively new biometric trait, Finger-Knuckle-Print (FKP) plays a vital role in establishing a personal authentication system in modern society due to its rich discriminative features, low time cost in image capture and user-friendliness. However, most existing KFP descriptors are hand-crafted and fail to work well with limited training samples. In this paper, we propose a feature learning method for few-shot FKP recognition by jointly learning compact multi-view hash codes (JLCMHC) of a FKP image. We first form the multi-view data vectors (MVDV) to exploit the multiple feature-specific information from a FKP image. Then, we learn a feature projection to encode the MVDV into compact binary codes in an unsupervised manner, where 1) the variance of the learned feature codes on each view is maximized and 2) the difference of the inter-view binary codes is enlarged, so that the redundant information in MVDV is reduced and more informative features can be obtained. Lastly, we pool the binary codes into block-wise statistics features as the final descriptor for FKP representation and recognition. Experimental results on the existing benchmark FKP databases clearly show that the JLCMHC method outperforms the state-of-the-art FKP descriptors. [ABSTRACT FROM AUTHOR]

• Proposing Cluster-wise Unsupervised Hashing to solve effective and efficient large-scale retrieval across modalities. • Utilizing the multi-view clustering to project the multi-modal data into its low-dimensional space, saving the diversity. • The proposed cluster-wise prototype making different data points in the same cluster having the same binary codes. • No much loss of information during transforming the real-valued data into binary codes based on the cluster-wise prototype. • Designing a discrete optimization framework to directly learn the unified binary codes for heterogeneous modalities. Cross-modal hashing similarity retrieval plays dual roles across various applications including search engines and autopilot systems. More generally, these methods also known to reduce the computation and memory storage in a training scheme. The key limitation of current methods are that: (i) they relax the discrete constrains to solve the optimization problem which may defeat the model purpose, (ii) projecting heterogenous data into a latent space may encourage to loss the diverse representations in such data, (iii) transforming real-valued data point to the binary codes always resulting in a loss of information and producing the suboptimal continuous latent space. In this paper, we propose a novel framework to project the original data points from different modalities into its own low-dimensional latent space and finds the cluster centroid points in its a low-dimensional space, using Cluster-wise Unsupervised Hashing (CUH). In particular, the proposed clustering scheme aims to jointly learns the compact hash codes and the corresponding linear hash functions. A discrete optimization framework is developed to learn the unified binary codes across modalities under of the guidance cluster-wise code-prototypes. Extensive experiments over multiple datasets demonstrate the effectiveness of our proposed model in comparison with the state-of-the-art in unsupervised cross-modal hashing tasks. [ABSTRACT FROM AUTHOR]

• Filter selection and model fine-tuning are integrated into a single end-to-end trainable framework. • Adaptive compression ratio and multi-layer compression. • Good generalization ability. Channel pruning is an important method to speed up CNN model's inference. Previous filter pruning algorithms regard importance evaluation and model fine-tuning as two independent steps. This paper argues that combining them into a single end-to-end trainable system will lead to better results. We propose an efficient channel selection layer, namely AutoPruner, to find less important filters automatically in a joint training manner. Our AutoPruner takes previous activation responses as an input and generates a true binary index code for pruning. Hence, all the filters corresponding to zero index values can be removed safely after training. By gradually erasing several unimportant filters, we can prevent an excessive drop in model accuracy. Compared with previous state-of-the-art pruning algorithms (including training from scratch), AutoPruner achieves significantly better performance. Furthermore, ablation experiments show that the proposed novel mini-batch pooling and binarization operations are vital for the success of model pruning. [ABSTRACT FROM AUTHOR]

• We propose a discrete deep matrix factorization to learn unified hash codes which can capture more complex structure within heterogeneous data, resulting in more representative hashing codes. • We further introduce a linear classification error term to make the learned unified hashing codes discriminative. • We directly conduct discrete optimization which can reduce the quantization error. Matrix factorization has been utilized for the task of cross-view hashing, where basis functions are learned to map data from different views to the same hamming embedding. It is possible that the basis functions between the hamming embedding and the original data matrix contain rather complex hierarchical information, which existing work can not capture. In addition, previous work employs relaxation technique in the matrix factorization based hashing which may lead to large quantization error. To address these issues, this paper presents a novel Supervised Discrete Deep Matrix Factorization (SDDMF) for cross-view hashing. We introduce deep matrix factorization so that SDDMF is able to learn a set of hierarchical basis functions and unified binary codes from different views. In addition, a classification error term is incorporated into the objective to learn discriminative binary codes. We then employ a linearization technique to directly optimize the discrete constraints which can significantly reduce the quantization error. Experimental results on three standard datasets with image-text modalities verify that SDDMF significantly outperforms several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• Our method adapts the LRR model into a new variant, based on which the learned correlation matrix could be designed into a space-and-time saving formula for data semantics. • To tackle the discrete graph hashing, we presents a new learning method, i.e., transforms the original optimization problem into three subproblems by means of surrogate variables, and most importantly each subproblem is addressed with a closed-form solution, which makes the whole hashing learning converge within dozens of iterations. • Experiments on four datasets demonstrate the advantages of our method over several state-of- the-art unsupervised hashing models including two recently proposed unsupervised deep hashing methods. Unsupervised semantic hashing should in principle keep the semantics among samples consistent with the intrinsic geometric structures of the dataset. In this paper, we propose a novel multiple stage unsupervised hashing method, named " Unsupervised Hashing based on the Recovery of Subspace Structures " (RSSH) for image retrieval. Specifically, we firstly adapt the Low-rank Representation (LRR) model into a new variant which treats the real-world data as samples drawn from a union of several low-rank subspaces. Then, the pairwise similarities are represented in a space-and-time saving manner based on the learned low-rank correlation matrix of the modified LRR. Next, the challenging discrete graph hashing is employed for binary hashing codes. Notably, we convert the original graph hashing model into an optimization-friendly formalization, which is addressed with efficient closed-form solutions for its subproblems. Finally, the devised linear hash functions are fast achieved for out-of-samples. Retrieval experiments on four image datasets testify the superiority of RSSH to several state-of-the-art hashing models. Besides, it's worth mentioning that RSSH, a shallow model, significantly outperforms two recently proposed unsupervised deep hashing methods, which further confirms its effectiveness. [ABSTRACT FROM AUTHOR]

• By employing the clustering technique and concatenating the substrings learnt by the hash functions in each cluster, the proposed method can model the complex relationship among the data and alleviate the effect brought from the boundary of the cluster. • An alternating optimization is developed to simultaneously discover the cluster structures of the data and learn the hash functions to preserve the relative positions of the data to each cluster center. • The experiments show that the proposed method is competitive to or better than other unsupervised hashing methods. Especially when learning the long codes in order to achieve the high search precision, the proposed method is obviously superior to the other methods. Hashing methods perform the efficient nearest neighbor search by mapping high-dimensional data to binary codes. Compared to projection-based hashing methods, hashing methods that adopt the clustering technique can encode the complex relationship of the data into binary codes. However, their search performance is affected by the boundary of the cluster. Two similar data points may be assigned to two different clusters and then encoded into two much different binary codes. In this paper, we propose a new hashing method based on the clustering technique and it can alleviate the effect from the cluster boundary. It is from an observation that the relative positions of any two close data points to each cluster center are close. An alternating optimization is developed to simultaneously discover the cluster structures of the data and learn the hash functions to preserve the relative positions of the data to each cluster center. To integrate the information in each cluster, the corresponding binary code of each data point is obtained by concatenating the substrings learnt by the hash functions in each cluster. The experiments show that our method is competitive to or better than the state-of-the-art hashing methods. [ABSTRACT FROM AUTHOR]

• We design a new objective function which not only considers the hamming distances between similar pairs, but also constrains the hamming distances between dissimilar pairs. • We further extend our method and give a kernelized version to address the circumstances in which the query subspaces have different dimensions. • Experiments on more public databases with different tasks are presented. The results show that our method outperforms all the state-of-the-art subspace search methods in both searching accuracy and efficiency. Subspace is widely used to represent objects under different viewpoints, illuminations, identities, and more. Due to the growing amount and dimensionality of visual contents, fast search in a large-scale database with high-dimensional subspaces is an important task in many applications, such as image retrieval, clustering, video retrieval, and visual recognition. This can be facilitated by approximate nearest subspace (ANS) search which requires effective subspace representation. All existing methods for this problem represent a subspace by a point in the Euclidean or the Grassmannian space before applying the approximate nearest neighbor (ANN) search. However, the efficiency of these methods is not guaranteed because the subspace representation step can be very time consuming when coping with high-dimensional data. Moreover, the subspace to point transforming process may cause subspace structural information loss which influences the search accuracy. In this paper, we present a new approach for hashing-based ANS search which can directly binarize a subspace without transforming it into a vector. The proposed method learns the binary codes for subspaces following a similarity preserving criterion, and simultaneously leverages the learned binary codes to train matrix classifiers as hash functions. Experiments on face and action recognition and video retrieval applications show that our method outperforms several state-of-the-art methods in both efficiency and accuracy. Moreover, we also compare our method with vector-based hashing methods. Results also show the superiority of our subspace matrix based search scheme. [ABSTRACT FROM AUTHOR]