Improved fast gauss transform and efficient kernel density estimation

By Changjiang Yang, Ramani Duraiswami, Nail A. Gumerov, Larry S. Davis. 2003

527 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

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

Prototypal Analysis and Prototypal Regression (2017, methodology): “For the Gaussian kernel, the time complexity for the inner product can be reduced to O(n1 + n2) via the fast Gauss transform (Greengard and Strain, 1991) or improved fast Gauss transform (Yang et al., 2003).”
Barnes-Hut-SNE (2013, methodology): “Prior work [6] has also used the fast Gaussian transform [9, 30] (a special case of a fast multipole method [19]) to speed up the computation of Gaussian N -body interactions.”
Geodesic graph cut for interactive image segmentation (2010, methodology): “We use the Fast Gauss Transform [32] to compute the foreground/background color models P l ( c ) (see Eq.”
Image and Video Matting: A Survey (2007, unknown): “…modeling • Learning and statistical methods • Appearance-based matching • Object and scene recognition • Face detection and recognition • Activity and gesture recognition • Image and Video Retrieval • Video analysis and event recognition • Medical Image Analysis • Robot Localization and Navigation”
Entropy Minimization for Shadow Removal (2009, background): “It has proved to be a very efficient algorithm in a variety of applications (Yang et al. 2003; Elgammal et al. 2003; Beatson and Greengard 1997).”
Classification using intersection kernel support vector machines is efficient (2008, methodology): “[29] who use the Fast Gauss Transform to build efﬁcient classiﬁers based on the Gaussian kernel.”
SHREC 2007 3 D Shape Retrieval Contest (2006, background): “They turn out to be useful for many applications, such as rapid prototyping or digital manifacturing.”
A Geodesic Framework for Fast Interactive Image and Video Segmentation and Matting (2007, methodology): “, PF (x) = Pr(x|F) Pr(x|F)+Pr(x|B) , where Pr(x|F) is the color PDF of ΩF , obtained via the fast kernel density estimation ([21]) (same process for the background PDF).”
A Review on Modal Clustering (2016, methodology): “Several adjustments to speed up the mean-shift procedures have been proposed (Yang et al., 2003a, Yang et al., 2003b), also specifically for some applications such as image segmentation (e.g., Lebourgeois et al., 2013).”
Geodesic Matting: A Framework for Fast Interactive Image and Video Segmentation and Matting (2009, methodology): “We estimate the probability density of the selected features, optical flow or filter responses in the previous examples and the 3D Lab color vector for most of the other examples presented in this paper, of the pixels covered by scribbles, via the fast kernel density estimation (Yang et al. 2003).”
Large-scale kernel machines (2007, methodology): “The core IFGT algorithm was first presented in (Yang et al., 2003) and its use in kernel machines was shown in (Yang et al.”
Accelerating t-SNE using tree-based algorithms (2014, methodology): “For Gaussian forces, the fast multipole approach is generally referred to as the fast Gauss transform (Greengard and Rokhlin, 1987; Yang et al., 2003).”
Quick Shift and Kernel Methods for Mode Seeking (2008, methodology): “In order to better appreciate the intrinsic efficiency advantages of each method, we use comparable vanilla implementations of the algorithms (in practice, one could use heuristics and advanced approximation techniques [23] to significantly accelerate the computation).”
Efficient mean-shift tracking via a new similarity measure (2005, methodology): “Section 4 describes the speedup of computing the similarity measure using the improved FGT. Section 5 gives some experimental results and Section 6 concludes the paper.”