The path assigned mean shift algorithm: A new fast mean shift implementation for colour image segmentation

Abstract

This paper presents a novel method for colour image segmentation derived from the mean shift theorem. When applied to colour image segmentation tasks, the path assigned mean shift algorithm performed 1.5 to 5 times faster than existing fast mean shift methods such as the Hierarchical 'Neighbourhood Consistency' FMS Method proposed by Zhang [1] with comparable results. The complexity of the new PAMS algorithm can be represented as O(φ 2 ) where φ represents the total number of unassigned points per iteration of the algorithm.

Dynamics of a mean-shift-like algorithm and its applications on clustering

“…In the generalized mean-shift algorithm (6), there is no restriction on Σ i . In addition, the fundamental, on which (6) is derived, is different from that of the standard MS algorithm.…”

“…The goal of the first example is to show the influence of σ on the number of equilibriums, convergence speed; and the aim of the other example is to show that the performance of the GMS algorithm composed by (6) and (7) is comparably favorable to the MS algorithm with homoscedastic or isotropic assumption, under the clustering procedure shown in Ta indicate that the GMS algorithm (7) has multiple equilibriums, and there is only one equilibrium in inset (c). So the change of σ influences the number of equilibriums.…”

“…The GMS algorithm, given by (6) and (7), has solutions in the convex hull of the data points, H convex (S).…”

“…The MS algorithm is an iterative procedure for locating the quadratic bound maxima of a density function given discrete data sampled from that function [2,3]. As a nonparametric clustering method [4], MS has proven particularly successful in image segmentation [5,6] and clustering [7,8] due to its nonparametric nature and the minimal user input. Recently, the algorithm has attracted attention in function by a finite mixture of kernel density estimates.…”

A Viterbi tracker for local features

“…The size of the clusters are determined by the Mean Shift kernel bandwidth (radius of circle). Accepted clusters (1,3,8,9) contain only a single descriptor from all three images n, n + 1, and n + 2.…”