Mean Shift Clustering Algorithm Research Articles

Investigation of methods for solving a large dimensional routing problem

In this paper, we researched methods for solving a large-scale vehicle routing problem. First, we provided the description of the problem that modern logistic companies encounter during the planning of transportations. As a solution for this problem, we used meta-heuristics and clustering algorithms. Existing studies and publications in the field of large-scale routing were analyzed, giving us the opportunity to assess the quality of the presented meta-heuristics and select those that lead to the best results. The definition of the routing problem is presented, as well as the description of requirements and limitations for the final solution. Then, in the main part of the work, we offer a brief overview of the Clark-Wright heuristic, which we use to build initial solutions. Also we give a brief description of Guided Local Search and Tabu Search algorithms. Both meta-heuristics help to improve the initial solution obtained using the local search algorithms, while avoiding falling into local minima. The K-Means and Mean-Shift clustering algorithms are described and used to speed up the calculation of the solution and eliminate unnecessary costs during routes optimization. Finally, to check the quality of the proposed solution, we developed software that implements the described meta-heuristics and clustering algorithms, and provides the routing results in the form of a route graph and its total length. The program was tested on the set of benchmarks of the Belgian roads, which were provided in the work of F. Arnold, M. Gendreau, K. Sorensen K. The results of comparison with the best solutions are provided along with the conclusion on the optimality of the results obtained. That allows us to assert that in less than ten minutes of solving a problem on each of the clusters, the developed software gives results that differ from the best known only by 5-7%. Further work will be aimed at improving the results by exchanging points between neighboring clusters.

Text detection and localization in natural scene images based on text awareness score

Text detection & localization plays an essential role in finding the textual information from natural scene images that can be used in robot navigation, license plate detection, and wearable applications. In this work, we present text detection and localization approach based upon a novel text awareness model that encompasses an improved fast edge preserving and smoothing Maximum Stable Extremal Region (FEPS-MSER) algorithm which uses the fast guided filter to separate the interconnected characters efficiently by removing the mixed pixels around the edges of blurred images. The fast guided filter takes less execution time as compared to other edge-smoothing filters. The combination of five independent and class determining facets namely stroke width deviation, 8-histogram of edge gradients, color variation, occupation ratio, and occupy rate convex area is proposed to differentiate between text and non-text components. The probability of a component to be text is based on Text Awareness Score (TAS) that is calculated by fusing these facets in Naive Bayes using the observation possibility and prior probability of text & non-text components. Naive Bayes classifier helps in accurate and fast determination of the text awareness score and thus helps in the classification of text & non-text components with the help of graph cut algorithm. The text components have been grouped by using the mean-shift clustering algorithm which is a non-parametric technique and does not require the initial knowledge of clusters. The proposed method achieves improved results concerning precision, recall, and f-measure on the ICDAR benchmark datasets for natural scene images.

Automatic cell nuclei segmentation and classification of cervical Pap smear images

Abstract Pathological examination of microscopic image of Pap smear slide remains the main method for cervical cancer diagnosis. The accurate segmentation and classification of images are two important phases of the analysis. Firstly, the Mean-Shift clustering algorithm is applied to obtain regions of interest (ROI) for cell nuclei segmentation. Then the flexible mathematical morphology is applied to split overlapped cell nuclei for better accuracy and robustness. For classification of the images, features based on shape, textural features based on color space and Gabor features are extracted and put together to obtain better classification performance. The optimal feature set is obtained by chain-like agent genetic algorithm (CAGA), P-value and maximum relevance-minimum multicollinearity (MRmMC). The proposed segmentation and classification methods were tested on 362 cervical Pap smear images. Experimental results showed that the cervical cell nuclei can be segmented by the proposed segmentation method with high effective segmentation results (Sensitivity: 94.25%±1.03% and Specificity 93.45%±1.14%). The feature selection method based on CAGA with Gabor features has the highest classification performance for normal, uninvolved and abnormal images (more than 96% accuracy). The proposed method can automatically and effectively segment cell nuclei of microscopic images. From the experimental results, Gabor features and feature selection based on CAGA are apparently helpful for improving the performance of classification.

Incorporating unsupervised learning into intrusion detection for wireless sensor networks with structural co-evolvability

Wireless sensor networks (WSNs) are vulnerable to many security threats because of the open and unreliable communication channels, the highly dynamic network structure as well as the decentralized management scheme. It is therefore, quite challenging to build an intrusion detection system that can detect various unknown attacks, reach better balance between detection rate and false alarm rate and increase the adaptivity to network dynamics, particularly for a resource-constraint WSN. In this paper, we proposed a knowledge-based intrusion detection strategy (KBIDS) to bridge the gap. We firstly used the Mean Shift Clustering Algorithm (MSCA), an unsupervised learning scheme to distinguish undefined abnormal patterns which reflect the abnormal behavior of a WSN being attacked from the normal context; then we employed a support vector machine to maximize the margin between abnormal and normal features so that the classification error can be minimized, which in turn to effectively enhance the detection accuracy; finally, we adopted a feature updating strategy to reflect network dynamics so that the system can co-evolve with the network change. Then, the validation of KBIDS in both network emulator and the real environment were conducted and analyzed. Results showed that KBIDS had achieved the highest detection rate and the lowest false alarm rate among several state-of-the-art intrusion models. In addition to that, we also conducted some parameter sensitivity analyses to help identifying the optimal configuration which can be used to parameterize KBIDS in real applications.

Similarity-based attribute weighting methods via clustering algorithms in the classification of imbalanced medical datasets

In the fields of pattern recognition and machine learning, the use of data preprocessing algorithms has been increasing in recent years to achieve high classification performance. In particular, it has become inevitable to use the data preprocessing method prior to classification algorithms in classifying medical datasets with the nonlinear and imbalanced data distribution. In this study, a new data preprocessing method has been proposed for the classification of Parkinson, hepatitis, Pima Indians, single proton emission computed tomography (SPECT) heart, and thoracic surgery medical datasets with the nonlinear and imbalanced data distribution. These datasets were taken from UCI machine learning repository. The proposed data preprocessing method consists of three steps. In the first step, the cluster centers of each attribute were calculated using k-means, fuzzy c-means, and mean shift clustering algorithms in medical datasets including Parkinson, hepatitis, Pima Indians, SPECT heart, and thoracic surgery medical datasets. In the second step, the absolute differences between the data in each attribute and the cluster centers are calculated, and then, the average of these differences is calculated for each attribute. In the final step, the weighting coefficients are calculated by dividing the mean value of the difference to the cluster centers, and then, weighting is performed by multiplying the obtained weight coefficients by the attribute values in the dataset. Three different attribute weighting methods have been proposed: (1) similarity-based attribute weighting in k-means clustering, (2) similarity-based attribute weighting in fuzzy c-means clustering, and (3) similarity-based attribute weighting in mean shift clustering. In this paper, we aimed to aggregate the data in each class together with the proposed attribute weighting methods and to reduce the variance value within the class. Thus, by reducing the value of variance in each class, we have put together the data in each class and at the same time, we have further increased the discrimination between the classes. To compare with other methods in the literature, the random subsampling has been used to handle the imbalanced dataset classification. After attribute weighting process, four classification algorithms including linear discriminant analysis, k-nearest neighbor classifier, support vector machine, and random forest classifier have been used to classify imbalanced medical datasets. To evaluate the performance of the proposed models, the classification accuracy, precision, recall, area under the ROC curve, κ value, and F-measure have been used. In the training and testing of the classifier models, three different methods including the 50–50% train–test holdout, the 60–40% train–test holdout, and tenfold cross-validation have been used. The experimental results have shown that the proposed attribute weighting methods have obtained higher classification performance than random subsampling method in the handling of classifying of the imbalanced medical datasets.

Automated method for extracting and analysing the rock discontinuities from point clouds based on digital surface model of rock mass

Often when rock discontinuities with complex distributions occur in steep terrain, it is difficult to rapidly survey and accurately measure their spatial distribution by traditional surveying and mapping methods. This paper presents a methodology for automated extraction of rock discontinuities from a point cloud and the resulting 3D digital surface model of the rock mass. First, feature planes of rock discontinuities are identified and classified using both their orientation and position in a double-nested Mean-shift Clustering Algorithm. Second, the points corresponding to feature planes are extracted using a Region Growth Algorithm and seed points. Finally, geological information is acquired by analysing the geometric features of the extracted sub-set of points from the point cloud. This approach can directly extract planar features associated with joints and it eliminates spurious points in a point cloud associated with objected such as vegetation. A case study of a rock slope along a highway is presented using the proposed method. A sensitivity analysis of relevant clustering parameters in the Mean-shift Clustering Algorithm is conducted to acquire their optimal values and to assess their robustness. The proposed method produces results that agree with the traditional survey methods and greatly improves the survey efficiency.

Clustering the Orion B giant molecular cloud based on its molecular emission.

Previous attempts at segmenting molecular line maps of molecular clouds have focused on using position-position-velocity data cubes of a single molecular line to separate the spatial components of the cloud. In contrast, wide field spectral imaging over a large spectral bandwidth in the (sub)mm domain now allows one to combine multiple molecular tracers to understand the different physical and chemical phases that constitute giant molecular clouds (GMCs). We aim at using multiple tracers (sensitive to different physical processes and conditions) to segment a molecular cloud into physically/chemically similar regions (rather than spatially connected components), thus disentangling the different physical/chemical phases present in the cloud. We use a machine learning clustering method, namely the Meanshift algorithm, to cluster pixels with similar molecular emission, ignoring spatial information. Clusters are defined around each maximum of the multidimensional Probability Density Function (PDF) of the line integrated intensities. Simple radiative transfer models were used to interpret the astrophysical information uncovered by the clustering analysis. A clustering analysis based only on the J = 1 - 0 lines of three isotopologues of CO proves suffcient to reveal distinct density/column density regimes (nH ~ 100 cm-3, ~ 500 cm-3, and > 1000 cm-3), closely related to the usual definitions of diffuse, translucent and high-column-density regions. Adding two UV-sensitive tracers, the J = 1 - 0 line of HCO+ and the N = 1 - 0 line of CN, allows us to distinguish two clearly distinct chemical regimes, characteristic of UV-illuminated and UV-shielded gas. The UV-illuminated regime shows overbright HCO+ and CN emission, which we relate to a photochemical enrichment effect. We also find a tail of high CN/HCO+ intensity ratio in UV-illuminated regions. Finer distinctions in density classes (nH ~ 7 × 103 cm-3 ~ 4 × 104 cm-3) for the densest regions are also identified, likely related to the higher critical density of the CN and HCO+ (1 - 0) lines. These distinctions are only possible because the high-density regions are spatially resolved. Molecules are versatile tracers of GMCs because their line intensities bear the signature of the physics and chemistry at play in the gas. The association of simultaneous multi-line, wide-field mapping and powerful machine learning methods such as the Meanshift clustering algorithm reveals how to decode the complex information available in these molecular tracers.

Intelligent recognition of dominant colors for Chinese traditional costumes based on a mean shift clustering method

Chinese traditional costumes have been recognized as one of the most influential sources for Western designers to obtain oriental inspirations and create Asian chic, especially the dominant colors and design style. In this paper, an effective color clustering method based on Mean shift clustering algorithm is developed for Chinese traditional costumes image. The proposed method consists of four steps: (1) costumes image acquisition, (2) costumes image denoising, (3) object segmentation, and (4) color clustering and dominant colors extraction. Firstly, a digital SLR (Single Lens Reflex) camera is used to capture the costumes images. Secondly, the sub-images in the three color channels are filtered by median filter separately. Thirdly, the filtered images are segmented based on the background color in the Lab color space, and the object costumes is separated from the background. Fourthly, the pixels of the costume image are classified into several clusters by Mean shift clustering algorithm, and the dominate colors are extracted from the classification results. The experimental results demonstrate that the proposed method can extract the dominant colors from costumes images with great accuracy when the bandwidth of Mean shift clustering algorithm is set as 0.05.

Comparative Analysis of Clustering Techniques for Movie Recommendation

Movie recommendation is a subject with immense ambiguity. A person might like a movie but not a very similar movie. The present recommending systems focus more on just few parameters such as Director, cast and genre. A lot of Power intensive methods such as Deep Convolutional Neural Network (CNN) has been used which demands the use of Graphics processors that require more energy. We try to accomplish the same task using lesser Energy consuming algorithms such as clustering techniques. In this paper, we try to create a more generalized list of similar movies in order to provide the user with more variety of movies which he/she might like, using clustering algorithms. We will compare how choosing different parameters and number of features affect the cluster's content. Also, compare how different algorithms such as K-mean, Hierarchical, Birch and mean shift clustering algorithms give a varied result and conclude which method will suit for which scenarios of movie recommendations. We also conclude on which algorithm clusters stray data points more efficiently and how different algorithms provide different advantages and disadvantages.

Weakly Supervised Foreground Segmentation Based on Superpixel Grouping

Image segmentation is of great significance to a variety of tasks in image processing and computer vision. Since fully unsupervised image segmentation is usually very hard in most cases, a task-oriented interactive segmentation approach becomes a popular solution. This paper proposes a weakly supervised image segmentation algorithm to extract foreground from a complex background relying only on a roughly predefined bounding-box. The algorithm integrates the Watershed algorithm and Mean-shift clustering algorithm to obtain reliable initial foreground and background labels for simple linear iterative clustering (SLIC) superpixels. Then, a synthetic superpixel grouping mechanism is proposed to group the remainder SLIC superpixels into foreground or background until the whole superpixels are completely grouped. The proposed algorithm reliefs the interactive information from users while maintaining the segmentation precision. Extensive experiments are performed, and the results indicate that the proposed algorithm can reliably segment the image foreground from the complex background with only a weakly supervision.

Sparse least squares support vector machines based on Meanshift clustering method

Aim to non-sparsity solution problem of least squares support vector machines and not zero support vector value, a Meanshift Clustering Algorithm to select training samples is proposed, in which keeps the samples with large contribution value and removes the samples with small contribution value. Sparse solution and effective classification model could be obtained. The experimental results on UCI test data set show that the proposed algorithm in this paper can effectively reduce the classification error and improve training efficiency by sparsing training samples of least squares support vector machines. The feasibility and effectiveness of proposed method are proved to be valid.

Multiprocessor architecture for real-time applications using mean shift clustering

Data mining and machine learning algorithms deal with large amount of data, which with the invention of cost-efficient devices has increased by massive amounts. Many algorithms of these domains are not part of real-time systems because of their computational complexity and large data on which they need to work. A lot of algorithms are being implemented on parallel processing systems like GPUs and FPGAs to achieve the desired speed. The purpose of this article is to provide parallel processing model of mean shift clustering algorithm, targeted to run on FPGA. The general model consists of multiple homogeneous processing entities (PEs) connected through a bus. These PEs work in collaborative working environment with each PE working independently and also communicating with its peers according to the requirements of algorithms. The proposed architecture is implemented on FPGA for one-dimensional data. The algorithm is tested on 99 images from segmentation evaluation database for different number of PEs and different number of fractional bits used to represent mean. The simplicity of algorithm resulted in utilizing only 10.31% of total device slice registers and 33% of total slice LUTs of Spartan 6 FPGA. The processing requirements for the proposed algorithm show that it can be used in real-time systems.

Federated Filter Algorithm with Positioning Technique Based on 3D Sensor

Somatosensory control based on 3D sensors has been widely used as a virtual reality and motion capture technology in large-screen human–computer interactive systems. However, many issues still exist in human–computer interactive systems, such as inaccuracies of target tracking and poor anti-jamming performance. A federated tracking filtering algorithm is proposed in this paper to address these problems. The algorithm first pretreats the original data with dynamic adaptive filtering and an unscented Kalman filter and then applies a mean shift clustering algorithm on the preprocessed data. Experiments show that the correct tracking rate can reach 95.3% when the measured points are 200, while the correct tracking rate of mean shift algorithm is 90.7%. Furthermore, the noise points are found to be almost completely filtered by comparing the depth image before and after filtering. The federated tracking filtering algorithm is shown to guarantee accurate target tracking and excellent anti-jamming performance, and can significantly improve the interactive experience.

A man-made object detection algorithm based on contour complexity evaluation

Man-made object detection is of great significance in both military and civil areas, such as search-and-rescue missions at sea, traffic signs recognition during visual navigation, and targets location in a military strike. Contours of man-made objects usually consist of straight lines, corner points, and simple curves. Motivated by this observation, a man-made object detection method is proposed based on complexity evaluation of object contours. After salient contours which keep the crucial information of objects are accurately extracted using an improved mean-shift clustering algorithm, a novel approach is presented to evaluate the complexity of contours. By comparing the entropy values of contours before/after sampling and linear interpolation, it is easy to distinguish between man-made objects and natural ones according to the complexity of their contours. Experimental results show that the presented method can effectively detect man-made objects when compared to the existing ones.

Image segmentation and classification of cytological cells based on multi-features clustering and chain splitting model

The diagnosis of pancreatic cancer is very important. The main method of diagnosis is based on pathological analysis of microscopic image of Pap smear slide. The accurate segmentation and classification of images are two important phases of the analysis. In this paper, we proposed a new automatic segmentation and classification method for microscopic images of pancreas. For the segmentation phase, firstly multi-features Mean-shift clustering algorithm (MFMS) was applied to localize regions of nuclei. Then, chain splitting model (CSM) containing flexible mathematical morphology and curvature scale space corner detection method was applied to split overlapped cells for better accuracy and robustness. For classification phase, 4 shape-based features and 138 textural features based on color spaces of cell nuclei were extracted. In order to achieve optimal feature set and classify different cells, chain-like agent genetic algorithm (CAGA) combined with support vector machine (SVM) was proposed. The proposed method was tested on 15 cytology images containing 461 cell nuclei. Experimental results showed that the proposed method could automatically segment and classify different types of microscopic images of pancreatic cell and had effective segmentation and classification results. The mean accuracy of segmentation is 93.46%±7.24%. The classification performance of normal and malignant cells can achieve 96.55%±0.99% for accuracy, 96.10%±3.08% for sensitivity and 96.80%±1.48% for specificity.

Mean-Shift segmentation and PDE-based nonlinear diffusion: toward a common variational framework for foreground/background document image segmentation

The presence of noise in images of degraded documents limits the direct application of segmentation approaches and can lead to the presence of a number of different artifacts in the final segmented image. A possible solution is the integration of a pre-filtering step which may improve the segmentation quality through the reduction of such noise. This study demonstrated that combining the Mean-Shift clustering algorithm and the tensor-driven diffusion process into a joint iterative framework produced promising results. For instance, this framework generates segmented images with reduced edge and background artifacts when compared to results obtained after applying each method separately. This improvement is explained by the mutual interaction of global and local information, introduced, respectively, by the Mean-Shift and the anisotropic diffusion. Another point of note is that the anisotropic diffusion process smoothed images while preserving edge continuities. The convergence of this framework was defined automatically under a stopping criterion not previously defined when the diffusion process was applied alone. To obtain a fast convergence, the common framework utilizes the speedup algorithm of the Fukunaga and Hostetler Mean-Shift formulation already proposed by Lebourgeois et al. (International Conference on Document Analysis and Recognition (ICDAR), pp 52–56, 2013). This new variant of the Mean-Shift algorithm produced similar results to the original one, but ran faster due to the application of the integral volume. The first application of this framework was document ink bleed-through removal where noise is stemmed from the interference of the verso side on the recto side, thus perturbing the legibility of the original text. Other categories of images could also be subjected to the proposed framework application.

PLDA-based mean shift speakers' short segments clustering

This paper extends upon a previous work using Mean Shift algorithm to perform speaker clustering on i-vectors generated from short speech segments. In this paper we examine the effectiveness of probabilistic linear discriminant analysis (PLDA) scoring as the metric of the mean shift clustering algorithm in the presence of different numbers of speakers. Our proposed method, combined with k-nearest neighbors (kNN) for bandwidth estimation, yields better and more robust results in comparison to the cosine similarity with fixed neighborhood bandwidth for clustering segments of large numbers of speakers. In the case of 30 speakers, we achieved significant improvement in cluster and speaker purity with the PLDA-based mean shift algorithm compared to the cosine-based baseline system.

Water Region Extraction in Thermal and RGB Sequences Using Spatiotemporally-Oriented Energy Features

Although the concept of Regions Of Interest (ROI) is known in video analysis, the ROI extraction problem has been hardly addressed in maritime surveillance, particularly for vessel detection and tracking. A video captured by a maritime surveillance camera may contain irrelevant regions, such as shorelines, bridges and piers. As a result, non-relevant moving objects (e.g. cars moving along the shorelines) can be misleadingly detected by the vessel or ship surveillance system. This paper proposes a robust water region extraction method based on spatiotemporallyoriented energy features in combination with a mean shift clustering algorithm. The method targets not only the conventional RGB surveillance data, but also data from thermal cameras. Experimental results reveal that the pixel-wise water segmentation recall is 95.23% on average for the RGB images and 94.29% on average for the thermal images, even in the presence of islands or other complex shoreline shapes. The measured average precisions are 93.88% and 95.41% for the RGB and thermal datasets, respectively.

Design and evaluation of an ambient intelligence testbed for improving quality of life

Ambient intelligence (AmI) deals with a new world of ubiquitous computing devices, where physical environments interact intelligently and unobtrusively with people. AmI environments can be diverse, such as homes, offices, meeting rooms, schools, hospitals, control centers, vehicles, tourist attractions, stores, sports facilities, and music devices. In this paper, we present the design and implementation of a testbed for AmI using Raspberry Pi mounted on Raspbian OS. We analyse the optimised link state routing (OLSR) and wired equivalent privacy (WEP) protocol in an indoor scenario, and mean shift clustering algorithm considering sensing data. For evaluation we considered throughput, delay and jitter metrics, and respiratory rate and heart rate metrics. The experimental and simulation results show that the nodes in the testbed were communicating smoothly and the mean shift clustering algorithm have a good performance.

An Adaptive Intrusion Detection Method for Wireless Sensor Networks

Current intrusion detection systems for Wireless Sensor Networks (WSNs) which are usually designed to detect a specific form of intrusion or only applied for one specific type of network structure has apparently restrictions in facing various attacks and different network structures. To bridge this gap, based on the mechanism that attacks are much likely to be deviated from normal features and from different shapes of aggregations in feature space, we proposed a knowledge based intrusion detection strategy (KBIDS) to detect multiple forms of attacks over different network structure. We firstly, in the training stage, used a modified unsupervised mean shift clustering algorithm to discover clusters in network features. Then the discovered clusters were classified as an anomaly if they had a certain amount of deviation from the normal cluster captured at the initial stage where no attacks could occur at all. The training data combined with a weighted support vector machine were then used to build the decision function that was used to flag network behaviors. The decision function was updated periodically after training by merging newly added network features to adapt network variability as well as to achieve time efficiency. During network running, each node uniformly captured their status as feature vector at certain interval and forwarded them to the base station on which the model was deployed and run. Using this way, our model can work independently of network structure in both detection and deployment. The efficiency and adaptability of the proposed method have been tested and evaluated by simulation experiments deployed on QualNet. The simulations were conducted as a full-factorial experiment in which all combinations of three forms of attacks and two types of WSN structures were tested. Results demonstrated that the detection accuracy and network structure adaptability of the proposed method outperforms the state-of-the-art intrusion detection methods for WSN.