Mean shift based clustering in high dimensions: a texture classification example

By Bogdan Georgescu, Ilan Shimshoni, Peter Meer. Proceedings Ninth IEEE International Conference on Computer Vision, 2003

500 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/c41bf09e692ae81b7f81e8c657303b30a4d807b5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Rapid object indexing using locality sensitive hashing and joint 3D-signature space estimation (2006, methodology): “To find the best values for the K and L, we have employed a method similar to [12].”
Automated Mapping of Phenotype Space with Single-Cell Data (2016, background): “Generally, those algorithms rely on kernel density estimation (KDE), the performance of which deteriorates in multidimensional data due to the sparseness of sample distribution 6 .”
Mid-level Visual Element Discovery as Discriminative Mode Seeking (2013, methodology): “While approximations exist [9], they rely on approximate nearest neighbor algorithms, which work for low-dimensional spaces ( 48 dimensions in [9]), but empirically we have found poor performance in HOG feature space (> 2000 dimensions).”
Sketch2Photo: internet image montage (2009, methodology): “Mean shift clustering [Georgescu et al. 2003] is employed to find clusters in feature space.”
Region Covariance: A Fast Descriptor for Detection and Classification (2006, methodology): “We compare our results with the results reported in [15] in Table 1.”
Locality-sensitive hashing scheme based on p-stable distributions (2004, background): “This is due to the fact that, in many cases, approximate nearest neighbor is almost as good as the exact one; in particular, if the distance measure accurately captures the notion of user quality, then small differences in the distance should not matter.”
From BoW to CNN: Two Decades of Texture Representation for Texture Classification (2018, methodology): “Many texture representation approaches exploit the Brodatz database for evaluations [99, 125, 163, 187, 190, 231], however in most cases the entire database is not utilized, except in some recent studies [67, 110, 132, 182, 263].”
Fast image search for learned metrics (2008, methodology): “Data-dependent variants of LSH have been proposed: the authors of [11] select partitions based on where data points are concentrated, while in [23] boosting is used to select feature dimensions that are most indicative of similarity in the parameter space.”
Creating efficient codebooks for visual recognition (2005, unknown): “Mean shift has been used in a similar context [11], but it can not be used directly with natural images because informative parts seldom coincide with well-separated density modes.”
How do humans sketch objects? (2012, methodology): “Instead, we use variable-bandwidth mean-shift clustering with locality sensitive hashing to speed up the underlying nearest-neighbor search problem [Georgescu et al. 2003].”
Simplifying Mixture Models Through Function Approximation (2006, background): “Therefore, mean shift has a time complexity of [14] and can be expensive on large data sets.”
Fast pose estimation with parameter-sensitive hashing (2003, methodology): “The idea of using approximate NN has been mentioned in previous work for object or texture classification [3, 10 ], and for some estimation tasks [13, 1]. However to our knowledge no experiments using recent algorithms for estimation tasks have been conducted.”
Skinning mesh animations (2005, methodology): “Consequently, we recommend the approach taken by [Georgescu et al. 2003], wherein localitysensitive hashing (LSH) is used for fast approximate nearest neighbor searches, and we use their publicly available mean shift implementation (see reference for website).”