The variable bandwidth mean shift and data-driven scale selection

Abstract

We present two solutions for the scale selection problem in computer vision. The first one is completely nonparametric and is based on the the adaptive estimation of the normalized density gradient. Employing the sample point estimator, we define the Variable Bandwidth Mean Shift, prove its convergence, and show its superiority over the fixed bandwidth procedure. The second technique has a semiparametric nature and imposes a local structure on the data to extract reliable scale information. The local scale of the underlying density is taken as the bandwidth which maximizes the magnitude of the normalized mean shift vector. Both estimators provide practical tools for autonomous image and quasi real-time video analysis and several examples are shown to illustrate their effectiveness.[15], which was proven to be superior to least squares cross validation and biased cross-validation [ll], [lG, ,

Visual MRI: Merging information visualization and non-parametric clustering techniques for MRI dataset analysis

“…As we will see in the following, setting the kernel bandwidth value could require a strong effort by the user: therefore, strategies to automatically determine the kernel bandwidth have been recently studied [25,32].…”

“…Roughly speaking, there are two main approaches addressing the adaptive bandwidth selection: (i) the non-parametric MS [32], and (ii) the Gaussian MS bandwidth estimators [25]. The first one [32] is based on the minimization of the mean integrated squared error (MISE) by inserting a plug-in rule into the MS clustering strategy [33].…”

“…The first one [32] is based on the minimization of the mean integrated squared error (MISE) by inserting a plug-in rule into the MS clustering strategy [33]. In general, plug-in rules consist in "plugging in" bandwidth estimates which minimize different kind of estimation errors evaluated on the entire pool of data [34].…”

Unsupervised Segmentation of Head Tissues from Multi-modal MR Images for EEG Source Localization

“…In our experiments below, we used the following multi-tissue segmentation algorithms (MTSAs): our own BAMS, AMS, FAST, FCM_S, and kmeans. Hereinafter, these four instantiations of the HSA are denoted HSA-BAMS, HSA-AMS, HSA-FAST, HSA-FCM_S, and HSA-k-means, Bayesian Adaptive Mean Shift BAMS is a variation on the AMS [22,23] segmentation method originally proposed by Mayer and Greenspan [18] for brain tissue segmentation in MR images. Mayer and Greenspan define the adaptive kernel bandwidth for the mean-shift algorithm in terms of the distance between the current feature point and its kth nearest neighbor.…”

Image Segmentation by Nonparametric Color Clustering

“…In the related works, however, the local bandwidth (or local scale) and structure information of neighborhood around individual sample are not considered adequately. In [7], the algorithms only take into account the difference of local bandwidth. The other algorithms do not involve the information of both local bandwidth and the anisotropy of sample's neighborhood in the feature space.…”