China
Africa
Explore chapters and articles related to this topic
Unsupervised Learning for Large Scale Data
Published in Ioannis S. Triantafyllou, Mangey Ram, Statistical Modeling of Reliability Structures and Industrial Processes, 2023
Petros Barmpas, Sotiris Tasoulis, Aristidis G. Vrahatis, Panagiotis Anagnostou, Spiros Georgakopoulos, Matthew Prina, José Luis Ayuso-Mateos, Jerome Bickenbach, Ivet Bayes, Martin Bobaki, Francisco Félix Caballero, Somnath Chatterji, Laia Egea-Cortés, Esther García-Esquinas, Matilde Leonardi, Seppo Koskinen, Ilona Koupil, Andrzej Pająk, Martin Prince, Warren Sanderson, Sergei Scherbov, Abdonas Tamosiunas, Aleksander Galas, Josep MariaHaro, Albert Sanchez-Niubo, Vassilis P. Plagianakos, Demosthenes Panagiotakos
Finally, in density-based clustering, a cluster is a set of data objects spread in the data space over a contiguous region of high density of objects. Density-based clusters are separated from each other by contiguous regions of low density of objects. Data objects located in low-density regions are typically considered noise or outliers. Density-based clustering algorithms are able to discover arbitrary-shaped clusters, but usually suffer from increased computational costs, preventing them to be scalable. DensityPeaks (Rodriguez and Laio 2014) algorithm is a novel density-based approach. Similar to the K-medoids method, it has its basis only in the distance between data points. Like DBSCAN (Hahsler et al. 2017) and the mean-shift process, it can detect non-spherical clusters and automatically find the correct number of clusters. As in the mean-shift method, the cluster centers are defined as local maxima in the density of data points. However, unlike the mean-shift method, this procedure does not require embedding the data in a vector space and maximizing explicitly the density field for each data point. The algorithm assumes that cluster centers are surrounded by regions with lower local density and are relatively far away from higher local density points. For each data point, the algorithm computes two measures: its local density and its distance from samples of higher density. Both these measures depend exclusively on the intervals between data points, which are considered to satisfy the triangular inequality.
Applications of Computer Vision
Published in Manas Kamal Bhuyan, Computer Vision and Image Processing, 2019
So, mean shift tracking is a kernel-based tracking, and the tracking is based on feature-space analysis. For example, the appearance of an object can be characterized using histograms, and tracking can be done based on these histograms. It is hard to specify an explicit 2D parametric motion model to track non-rigid objects (like a walking person). Appearances of non-rigid objects can sometimes be modeled with color distributions. Mean shift is basically a type of iterative clustering algorithm, which can provide the density gradient estimates irrespective of the prior information of number and shape of clusters. Mean shift algorithm consists of following iterative steps for positioning the objects: Initialization of the position of a fix-sized search window.Finding of the average position in the search window.Estimating center of the window at the average position, estimated in Step 2.Repeat Steps 2 and 3 until the average position changes less than a prior-set threshold. In this way, convergence is achieved.
A Survey of Partitional and Hierarchical Clustering Algorithms
Published in Charu C. Aggarwal, Chandan K. Reddy, Data Clustering, 2018
Chandan K. Reddy, Bhanukiran Vinzamuri
Mean shift clustering [7] is a popular nonparametric clustering technique which has been used in many areas of pattern recognition and computer vision. It aims to discover the modes present in the data through a convergence routine. The primary goal of the mean shift procedure is to determine the local maxima or modes present in the data distribution. The Parzen window kernel density estimation method forms the basis for the mean shift clustering algorithm. It starts with each point and then performs a gradient ascent procedure until convergence. As the mean shift vector always points toward the direction of maximum increase in the density, it can define a path leading to a stationary point of the estimated density. The local maxima (or modes) of the density are such stationary points. This mean shift algorithm is one of the widely used clustering methods that fall into the category of mode-finding procedures.
An integrated damage modeling and assessment framework for overhead power distribution systems considering tree-failure risks
Published in Structure and Infrastructure Engineering, 2022
If a sub-image is classified as a tree, its center is recognised as the tree crown center point. A tree crown or its part can be contained in several cropped sub-images. Thus, it can be repeatedly recognised by the classifier generating multiple center points. To address this issue, mean shift clustering can be employed to pinpoint the crown centroid. Mean shift represents a non-parametric mode clustering method treating points in feature space corresponding to underlying distributions (Fukunaga & Hostetler, 1975). It can be applied to cluster recognised center points into several groups. Each point group belongs to a tree, and its center is considered as the center of a tree crown. Figure 6 shows detected trees along two spans. The yellow bounding boxes do not represent the crown sizes. Since data on tree genus is unavailable, the tree genus is not detected by computer vision technique but assigned according to statistical data on trees of Connecticut introduced in Section 3.3. As shown in Figure 6, the detection accuracy for open trees is higher than for those in stands, as the edges of trees in stands might be blurred.
Nyström-based spectral clustering using airborne LiDAR point cloud data for individual tree segmentation
Published in International Journal of Digital Earth, 2021
Yong Pang, Weiwei Wang, Liming Du, Zhongjun Zhang, Xiaojun Liang, Yongning Li, Zuyuan Wang
The traditional voxel-based approach transforms the point cloud into cube voxels with a given resolution. Its main advantage is the convenient access to the neighbourhood of each voxel point. However, a small resolution is usually chosen to obtain a forest structure that is highly close to the original LiDAR point cloud, which limited compression of the data volume. The super voxel based method provides alternatively a more flexible and efficient approach to constructing a tighter voxel space. Mean shift is a clustering algorithm that groups points by iteratively shifting each point towards the density maxima within a kernel. It is a robust method with fast and good results that does not need to assume the distribution of data points or the number of clusters. On account of the advantages of mean shift method, we used it to complete the voxelization procedure. As shown in Figure 3, each voxel was represented by a cluster obtained based on the mean shift. The voxel position was determined by the centre point and the voxel weight is equal to the number of points within it. As a result, mean shift voxelization reduced the amount of data by approximately a factor of 10 (Figure 3).
Intelligent recognition of dominant colors for Chinese traditional costumes based on a mean shift clustering method
Published in The Journal of The Textile Institute, 2018
Le Xing, Jie Zhang, Hui’e Liang, Zhongjian Li
The pixels in the object costume are classified by the mean shift clustering method based on their color features in CIE Lab space. Several colors including the ground colors, adjunctive colors, and ornamental colors, are separated out automatically from costume image. Thus, the dominant colors can be extracted from these clustering colors. The bandwidth h is an important parameter for mean shift clustering method, which has greatly influence on clustering performance and computing time. In this section, the value of bandwidth is set as 0.05. The detail discussion of this parameter will be conducted in the Section 5.1. The clustering results of different colors as the clustering category labels are shown in Figure 7(a). The clustering result in RGB space is shown in Figure 7(b).