Nearest-neighbor Graph Research Articles

A Novel Local Density Hierarchical Clustering Algorithm Based on Reverse Nearest Neighbors

Clustering is widely used in data analysis, and density-based methods are developed rapidly in the recent 10 years. Although the state-of-art density peak clustering algorithms are efficient and can detect arbitrary shape clusters, they are nonsphere type of centroid-based methods essentially. In this paper, a novel local density hierarchical clustering algorithm based on reverse nearest neighbors, RNN-LDH, is proposed. By constructing and using a reverse nearest neighbor graph, the extended core regions are found out as initial clusters. Then, a new local density metric is defined to calculate the density of each object; meanwhile, the density hierarchical relationships among the objects are built according to their densities and neighbor relations. Finally, each unclustered object is classified to one of the initial clusters or noise. Results of experiments on synthetic and real data sets show that RNN-LDH outperforms the current clustering methods based on density peak or reverse nearest neighbors.

Hierarchical Clustering-Based Graphs for Large Scale Approximate Nearest Neighbor Search

This paper presents a novel approach to perform fast approximate nearest neighbors search in high dimensional data, using a nearest neighbor graph created over large collections. This graph is created based on the fusion of multiple hierarchical clustering results, where a minimum-spanning-tree structure is used to connect all elements in a cluster. We propose a novel search technique to guide the navigation on the graph without computing exhaustively the distances to all neighbors in each step of the search, just to those in the direction of the query. The objective is to determine the nearest point to the query with a few number of distance calculations. We experimented in three datasets of 1 million SIFT, GIST, and GloVe features. Results show better speedups than another graph-based technique, and competitive speedups at high recall values when compared to classic and recent state-of-the-art techniques.

Edge integrity of nearest neighbor graphs and separator theorems

We give upper bounds for the edge integrity and the vertex integrity of k-type nearest neighbor graphs in Rd. For n the order of a graph, examples are constructed to show that the bounds are best possible in the parameters n and k. The results follow from decompositions that are shown for certain graphs that possess small edge or vertex separators. A vertex separator theorem for nearest neighbor graphs is known, and an edge separator theorem is shown. We also apply the decompositions to obtain upper bounds for the edge integrity of planar graphs, graphs of fixed genus, and trees.

Towards Automatic Segmentation and Recognition of Multiple Precast Concrete Elements in Outdoor Laser Scan Data

To date, to improve construction quality and efficiency and reduce environmental pollution, the use of precast concrete elements (PCEs) has become popular in civil engineering. As PCEs are manufactured in a batch manner and possess complicated shapes, traditional manual inspection methods cannot meet today’s requirements in terms of production rate of PCEs. The manual inspection of PCEs needs to be conducted one by one after the production, resulting in the excessive storage of finished PCEs in the storage yards. Therefore, many studies have proposed the use of terrestrial laser scanners (TLSs) for the quality inspection of PCEs. However, all these studies focus on the data of a single PCE or a single surface of PCE, which is acquired from a unique or predefined scanning angle. It is thus still inefficient and impractical in reality, where hundred types of PCEs with different properties may exist. Taking this cue, this study proposes to scan multiple PCEs simultaneously to improve the inspection efficiency by using TLSs. In particular, a segmentation and recognition approach is proposed to automatically extract and identify the different types of PCEs in a large amount of outdoor laser scan data. For the data segmentation, 3D data is first converted into 2D images. Image processing is then combined with radially bounded nearest neighbor graph (RBNN) algorithm to speed up the laser scan data segmentation. For the PCE recognition, based on the as-designed models of PCEs in building information modeling (BIM), the proposed method uses a coarse matching and a fine matching to recognize the type of each PCE data. To the best of our knowledge, no research work has been conducted on the automatic recognition of PCEs from a million or even ten million of the outdoor laser scan points, which contain many different types of PCEs. To verify the feasibility of the proposed method, experimental studies have been conducted on the PCE outdoor laser scan data, considering the shape, type, and amount of PCEs. In total, 22 PCEs including 12 different types are involved in this paper. Experiment results confirm the effectiveness and efficiency of the proposed approach for automatic segmentation and recognition of different PCEs.

Transfer Linear Subspace Learning for Cross-Corpus Speech Emotion Recognition

Speech emotion recognition has received an increasing interest in recent years, which is often conducted on the assumption that speech utterances in training and testing datasets are obtained under the same conditions. However, in reality, this assumption does not hold as the speech data are often collected from different devices or environments. Hence, there exists discrepancy between the training and testing data, which will have an adverse effect on recognition performance. In this paper, we examine the problem of cross-corpus speech emotion recognition. To address it, we present a novel transfer linear subspace learning (TLSL) framework to learn a common feature subspace for source and target datasets. In TLSL, a nearest neighbor graph algorithm is used to measure the similarity between different corpora, and a feature grouping strategy is introduced to divide the emotional features into two categories, i.e., high transferable part (HTP) versus low transferable part (LTP). To explore the proposed TLSL with different scenarios, we propose two kinds of TLSL approaches, called transfer unsupervised linear subspace learning (TULSL) and transfer supervised linear subspace learning (TSLSL), and provide the corresponding solutions for the optimization problems. Extensive experiments on several benchmark datasets validate the effectiveness of TLSL for cross-corpus speech emotion recognition.

Recursive Nearest Neighbor Graph Partitioning for Extreme Multi-Label Learning

Recursive Nearest Neighbor Graph Partitioning for Extreme Multi-Label Learning

Graph-based particular object discovery

Severe background clutter is challenging in many computer vision tasks, including large-scale image retrieval. Global descriptors, which are popular due to their memory and search efficiency, are especially prone to corruption by such a clutter. Eliminating the impact of the clutter on the image descriptor increases the chance of retrieving relevant images and prevents topic drift due to actually retrieving the clutter in the case of query expansion. In this work, we propose a novel salient region detection method. It captures, in an unsupervised manner, patterns that are both discriminative and common in the dataset. Saliency is based on a centrality measure of a nearest neighbor graph constructed from regional CNN representations of dataset images. The proposed method exploits recent CNN architectures trained for object retrieval to construct the image representation from the salient regions. We improve particular object retrieval on challenging datasets containing small objects.

Feature-Based Deformable Registration Using Minimal Spanning Tree for Prostate MR Segmentation

Automatic and accurate segmentation of the prostate is still a challenging task due to intensity inhomogeneity and complicated deformation of MR images. To tackle these problems with multi-atlas segmentation, in this paper, we propose a new metric for image registration and new descriptor for label fusion. First, to reduce the amount of edges in entropic graph, a modified $\alpha $ -mutual information ( $\alpha $ -MI) based on fast minimal spanning tree (MST) is implemented for deformable registration. Second, localized $\alpha $ -MI allowing for the spatial information is proposed with the stochastic gradient optimization, and the feature space is encoded by a sparse auto-encoder. Finally, a multi-scale descriptor utilizing local self-similarity is integrated into the patch-based label fusion to obtain final segmentation. Experiments were performed on two subsets of totally 46 T2-weighted prostate MR images from 46 patients. Compared to $\alpha $ -MI based on ${k}$ -nearest neighbor graph, the registration time of $\alpha $ -MI based on fast MST can be reduced by almost half. The median Dice overlap of registration using localized $\alpha $ -MI on one subset is shown to improve significantly from 0.725 to 0.764 ( $p=1.14\times 10^{-5}$ ), compared to using $\alpha $ -MI without the spatial information. The median Dice overlap of prostate segmentation using the proposed method on 20 testing images of another subset is 0.871, and the median Hausdorff distance is 8.013 mm, which demonstrate a comparable accuracy to state-of-the-art methods.

Local Graph Reconstruction for Parameter Free Unsupervised Feature Selection

Facing with the absence of supervised information to guide the search of relevant features and the grid-search of model/hyper-parameters, it is more preferred to develop parameter-free methods and avoid additional hyper-parameters tuning. In this paper, we propose a new simple and effective parameter-free unsupervised feature selection algorithm by minimizing the linear reconstruction weight between the nearest neighbor graphs constructed from all candidate features and each single feature. The obtained global optimal reconstruction weights actually select those features with highest relevance and lowest redundancy simultaneously. The experimental results on many benchmark data sets demonstrate that the proposed method outperforms many of the state-of-the-art unsupervised feature selection methods.

Laplacian regularized generalized Dirichlet mixture distribution for data clustering

The techniques for data clustering have been frequently applied in machine learning and computer vision, data mining. Among of these methods the Gaussian mixture model (GMM) has received considerable attention due to its flexibility, intuitiveness and ease of use. However, ignoring the non-Gaussian distribution, heavy computing burden for high dimension data, and the manifold structure of the data depress the performance of GMM. To overcome these limits, we propose the generalized Dirichlet mixture model with Laplacian regularization for clustering. The generalized Dirichle mixture model can fit both Gaussian and non-Gaussian type data, which is superior to the Gaussian mixture model. Based on the assumption that two data points are likely to be generated from the same Dirichlet component if they are sufficiently close on the manifold, we build a nearest neighbor graph to model the manifold structure and incorporate the manifold structure in a log-likelihood function of the Dirichlet mixture model as a regularization term. During the iteration, we employ the property of generalized Dirichle distribution to reduce the computation burden that decreases computation complexity. Extensive experimental results on four challenging databases demonstrate that the proposed algorithm performs favorably against the previous methods.

Exploring Rotation and Scale Invariant Features in Image Pla- giarism Detection Using Manifold-Ranking Algorithm

Plagiarism, as a crucial offense especially in academia, not only is well-known problem in text but also is becoming widespread in image. In this work, the performance of manifold-ranking, known as robust method among semi-supervised methods, has been investigated by using twelve different features. As its high performance is attributed to the quality of constructed graph, we applied robust k-regular nearest neighbor (k-RNN) graph in the framework of manifold-ranking based retrieval. Among all tested feature point detectors and descriptors, Root-SIFT, the feature point ones, due to it is invariant to an array of image transforms, is the most reliable feature for calculating image similarity. The database consisting of images from scientific papers containing four popular benchmark test images served to test these methods.

A Low Arithmetic-Degree Algorithm for Computing Proximity Graphs

In this paper, we study the problem of designing robust algorithms for computing the minimum spanning tree, the nearest neighbor graph, and the relative neighborhood graph of a set of points in the plane, under the Euclidean metric. We use the term “robust” to denote an algorithm that can properly handle degenerate configurations of the input (such as co-circularities and collinearities) and that is not affected by errors in the flow of control due to round-off approximations. Existing asymptotically optimal algorithms that compute such graphs are either suboptimal in terms of the arithmetic precision required for the implementation, or cannot handle degeneracies, or are based on complex data structures. We present a unified approach to the robust computation of the above graphs. The approach is a variant of the general region approach for the computation of proximity graphs based on Yao graphs, first introduced in Ref. 43 (A. C.-C. Yao, On constructing minimum spanning trees in [Formula: see text]-dimensional spaces and related problems, SIAM J. Comput. 11(4) (1982) 721–736). We show that a sparse supergraph of these geometric graphs can be computed in asymptotically optimal time and space, and requiring only double precision arithmetic, which is proved to be optimal. The arithmetic complexity of the approach is measured by using the notion of degree, introduced in Ref. 31 (G. Liotta, F. P. Preparata and R. Tamassia, Robust proximity queries: An illustration of degree-driven algorithm design, SIAM J. Comput. 28(3) (1998) 864–889) and Ref. 3 (J. D. Boissonnat and F. P. Preparata, Robust plane sweep for intersecting segments, SIAM J. Comput. 29(5) (2000) 1401–1421). As a side effect of our results, we solve a question left open by Katajainen27 (J. Katajainen, The region approach for computing relative neighborhood graphs in the [Formula: see text] metric, Computing 40 (1987) 147–161) about the existence of a subquadratic algorithm, based on the region approach, that computes the relative neighborhood graph of a set of points [Formula: see text] in the plane under the [Formula: see text] metric.

Visual tracking via Graph Regularized Kernel Correlation Filer and Multi-Memory Voting

Correlation filter based tracking approach has been an important branch of visual tracking. However, most correlation filter based trackers fail to work under occlusion due to their frame-by-frame model update strategy, and the tracking performance can be further enhanced by optimizing the energy equation. The target appearance during tracking is nearly moving on a manifold. So, the classification scores should be similar on the target manifold. K Nearest Neighbor graphs are constructed and the classification scores on the neighborhood are regularized to have similar values. Through the local score propagation on the graph, the learned Graph Regularized Kernel Correlation Filer can represent different appearances of the object. Furthermore, in the proposed Multi-Memory Voting scheme, occlusion problem is addressed by voting from multiple target snapshots in the memory pool. An extensive evaluation on two recent benchmarks shows that the proposed tracker achieves competitive performance compared to nine other state-of-the-art trackers.

A Segmentation Method for High Spatial Resolution Remote Sensing Images Based on the Fusion of Multifeatures

A novel method based on the fusion of spectral, texture, and shape features is proposed for the segmentation of high spatial resolution remote sensing images. The method uses the region merging idea to get the final segmentation result on the basis of initial segmentation. Texture features of the regions are obtained by the nonsubsampled contourlet transform. An integrated region merging criterion is built by combining the texture, spectral, and shape features. To ensure the efficiency of the method, the region adjacency graph and the nearest neighbor graph are used to maintain the adjacency relations in the region merging stage. The global optimization strategy is adopted to realize the image segmentation gradually. Two experiments are performed to verify the effectiveness of the proposed method. One experiment validates the influence of the merging criteria with different feature combinations on the segmentation results. Another experiment compares the effect of the method with the segmentation methods embedded in ENVI and eCognition. Experimental results show that the method can make full use of the features of the images to achieve accurate and efficient segmentations.

Semi-Supervised Maximum Discriminative Local Margin for Gene Selection

In the present study, we introduce a novel semi-supervised method called the semi-supervised maximum discriminative local margin (semiMM) for gene selection in expression data. The semiMM is a “filter” approach that exploits local structure, variance, and mutual information. We first constructed a local nearest neighbour graph and divided this information into within-class and between-class local nearest neighbour graphs by weighing the edge between the two data points. The semiMM aims to discover the most discriminative features for classification via maximizing the local margin between the within-class and between-class data, the variance of all data, and the mutual information of features with class labels. Experiments on five publicly available gene expression datasets revealed the effectiveness of the proposed method compared to three state-of-the-art feature selection algorithms.

RNN-DBSCAN: A Density-Based Clustering Algorithm Using Reverse Nearest Neighbor Density Estimates

A new density-based clustering algorithm, RNN-DBSCAN , is presented which uses reverse nearest neighbor counts as an estimate of observation density. Clustering is performed using a DBSCAN -like approach based on $k$ nearest neighbor graph traversals through dense observations. RNN-DBSCAN is preferable to the popular density-based clustering algorithm DBSCAN in two aspects. First, problem complexity is reduced to the use of a single parameter (choice of $k$ nearest neighbors), and second, an improved ability for handling large variations in cluster density (heterogeneous density). The superiority of RNN-DBSCAN is demonstrated on several artificial and real-world datasets with respect to prior work on reverse nearest neighbor based clustering approaches ( RECORD , IS-DBSCAN , and ISB-DBSCAN ) along with DBSCAN and OPTICS . Each of these clustering approaches is described by a common graph-based interpretation wherein clusters of dense observations are defined as connected components, along with a discussion on their computational complexity. Heuristics for RNN-DBSCAN parameter selection are presented, and the effects of $k$ on RNN-DBSCAN clusterings discussed. Additionally, with respect to scalability, an approximate version of RNN-DBSCAN is presented leveraging an existing approximate $k$ nearest neighbor technique.

Fast Approximate Nearest Neighbor Search via k-Diverse Nearest Neighbor Graph

Approximate nearest neighbor search is a fundamental problem and has been studied for a few decades. Recently graph-based indexing methods have demonstrated their great efficiency, whose main idea is to construct neighborhood graph offline and perform a greedy search starting from some sampled points of the graph online. Most existing graph-based methods focus on either the precise k-nearest neighbor (k-NN) graph which has good exploitation ability, or the diverse graph which has good exploration ability. In this paper, we propose the k-diverse nearest neighbor (k-DNN) graph, which balances the precision and diversity of the graph, leading to good exploitation and exploration abilities simultaneously. We introduce an efficient indexing algorithm for the construction of the k-DNN graph inspired by a well-known diverse ranking algorithm in information retrieval (IR). Experimental results show that our method can outperform both state-of-the-art precise graph and diverse graph methods.

All Near Neighbor GraphWithout Searching

Given a collection of n objects equipped with a distance function d(·, ·), the Nearest Neighbor Graph (NNG) consists in finding the nearest neighbor of each object in the collection. Without an index the total cost of NNG is quadratic. Using an index the cost would be sub-quadratic if the search for individual items is sublinear. Unfortunately, due to the so called curse of dimensionality the indexed and the brute force methods are almost equally inefficient. In this paper we present an efficient algorithm to build the Near Neighbor Graph (nNG), that is an approximation of NNG, using only the index construction, without actually searching for objects.

Conservation of a domestic metapopulation structured into related and partly admixed strains.

Preservation of genetic diversity is one of the most pressing challenges in the planetary boundaries concept. Within this context, we focused on genetic diversity in a native, unselected and highly admixed domesticated metapopulation. A set of 1,828 individuals from 60 different cattle breeds was analysed using a medium density SNP chip. Among these breeds, 14 Buša strains formed a metapopulation represented by 350 individuals, while the remaining 46 breeds represented the global cattle population. Genetic analyses showed that the scarcely selected and less differentiated Buša metapopulation contributed a substantial proportion (52.6%) of the neutral allelic diversity to this global taurine population. Consequently, there is an urgent need for synchronized maintenance of this highly fragmented domestic metapopulation, which is distributed over several countries without sophisticated infrastructure and highly endangered by continuous replacement crossing as part of the global genetic homogenization process. This study collected and evaluated samples, data and genomewide information and developed genome-assisted cross-border conservation concepts. To detect and maintain genetic integrity of the metapopulation strains, we designed and applied a composite test that combines six metrics based on additive genetic relationships, a nearest neighbour graph and the distribution of semiprivate alleles. Each metric provides distinct information components about past admixture events and offers an objective and powerful tool for the detection of admixed outliers. The here developed conservation methods and presented experiences could easily be adapted to comparable conservation programmes of domesticated or other metapopulations bred and kept in captivity or under some other sort of human control.

Low-Rank 2-D Neighborhood Preserving Projection for Enhanced Robust Image Representation.

2-D neighborhood preserving projection (2DNPP) uses 2-D images as feature input instead of 1-D vectors used by neighborhood preserving projection (NPP). 2DNPP requires less computation time than NPP. However, both NPP and 2DNPP use the L 2 norm as a metric, which is sensitive to noise in data. In this paper, we proposed a novel NPP method called low-rank 2DNPP (LR-2DNPP). This method divided the input data into a component part that encoded low-rank features, and an error part that ensured the noise was sparse. Then, a nearest neighbor graph was learned from the clean data using the same procedure as 2DNPP. To ensure that the features learned by LR-2DNPP were optimal for classification, we combined the structurally incoherent learning and low-rank learning with NPP to form a unified model called discriminative LR-2DNPP (DLR-2DNPP). By encoding the structural incoherence of the learned clean data, DLR-2DNPP could enhance the discriminative ability for feature extraction. Theoretical analyses on the convergence and computational complexity of LR-2DNPP and DLR-2DNPP were presented in details. We used seven public image databases to verify the performance of the proposed methods. The experimental results showed the effectiveness of our methods for robust image representation.