MDPE: A Very Robust Estimator for Model Fitting and Range Image Segmentation
Citations
169 citations
Cites methods from "MDPE: A Very Robust Estimator for M..."
...Indeed, we propose two novel robust techniques: the Two-Step Scale estimator (TSSE) and the Adaptive Scale Sample Consensus (ASSC) estimator....
[...]
...In this section, we investigate the behavior of several robust scale estimators that are widely used in computer vision community: showing some of the weaknesses of these scale estimation techniques....
[...]
...By employing TSSE in a RANSAC like procedure, we propose a highly robust estimator: Adaptive Scale Sample Consensus (ASSC) estimator....
[...]
...In the next section, we will propose a new robust estimator—Adaptive Scale Sample Consensus (ASSC) estimator, which can estimate the parameters and the scale simultaneously....
[...]
...Index Terms—Robust model fitting, random sample consensus, least-median-of-squares, residual consensus, adaptive least kth order squares, kernel density estimation, mean shift, range image segmentation, fundamental matrix estimation....
[...]
103 citations
Cites methods from "MDPE: A Very Robust Estimator for M..."
...In the computer vision literature, the latter approach dominates most applications [2, 4, 21, 25, 30] because applying RANSAC on individual simple geometric models has been well studied, and the algorithm complexity is much lower....
[...]
97 citations
Cites methods from "MDPE: A Very Robust Estimator for M..."
...RANSAC is performed in a randomized, hypothesize-andverify manner, which can yield good-quality estimates even when more than 50% of sample points are considered as gross errors [44], [45]....
[...]
90 citations
Cites background or methods from "MDPE: A Very Robust Estimator for M..."
...Instead of studying the distribution of N residuals per hypothesis as in [7] when trying to determine the threshold for inlier classification, we propose to study the distribution of M residuals for each data point xi....
[...]
...In [7] the authors propose a novel MDPE estimator (Maximal Density Power Estimator), which selects a hypothesis, whose corresponding density of residuals is maximal, with the mean close to zero....
[...]
...According to the result of [7], existing robust estimators are likely to fail in this case....
[...]
...The approach described here shares some features of the method proposed in [7], but differs in significant ways, which enable significant extensions to estimation of multiple models....
[...]
87 citations
References
23,396 citations
"MDPE: A Very Robust Estimator for M..." refers methods in this paper
...Experiments, presented next, show the MDPE is a very powerful method for data with a large percentage of outliers....
[...]
...RANSAC (Fischler and Rolles, 1981) applies criterion (1) into its optimization process and outputs the results with the highest number of data points within an error bound; The Least squares method uses criterion (2) as its objective function, but minimizes the residuals of all data points without…...
[...]
15,499 citations
11,727 citations
"MDPE: A Very Robust Estimator for M..." refers background or methods in this paper
...Experiment 4 The problem of the choice of window radius in the means shift, i.e., bandwidth selection, has been widely investigated during the past decades (Silverman, 1986; Wand and Jones, 1995; Comaniciu and Meer, 1999, 2002; Comaniciu et al., 2001)....
[...]
...Comaniciu and Meer (2002) suggested several techniques for the choice of window radius: (1) The optimal bandwidth should be the one that minimizes AMISE; (2) The choice of the bandwidth can be taken as the center of the largest operating range over which the same results are obtained for the same…...
[...]
...The proof of the convergence of the mean shift algorithm can be found in Comaniciu and Meer (1999, 2002)....
[...]
...Since its introduction by Fukunaga and Hostetler (1975), the mean shift method has been extensively exploited and applied in low level computer vision tasks (Cheng, 1995; Comaniciu and Meer, 1997, 1999, 2002) for its ease and efficiency....
[...]
10,526 citations
6,955 citations
"MDPE: A Very Robust Estimator for M..." refers background or methods in this paper
...In order to improve the statistical efficiency, a weighted least square procedure (Rousseeuw and Leroy, 1987, p. 202) can be carried out after the initial MDPE fit....
[...]
...Moreover, though there is a formal proof of high breakdown point (0.5, see Rousseeuw and Leroy, 1987, p. 125), this proof only applies to the exact LMedS and not the approximate method (using random sampling) that has to be used in practice....
[...]
...In the statistical literature (Huber, 1981; Rousseeuw and Leroy, 1987), there are a number of precise definitions of robustness and of robust properties: including the aforementioned “breakdown point”—which is an attempt to characterize the tolerance of an estimator to large percentages of outliers....
[...]
...The breakdown point of an estimator may be roughly defined as the smallest percentage of outlier contamination that can cause the estimator to produce arbitrarily large values (Rousseeuw and Leroy, 1987, p. 9)....
[...]
...Although the breakdown point in statistics is proved to be bounded by 0.5 (Rousseeuw and Leroy, 1987, p. 125), the proof shows that they require the robust estimator has a unique solution (more technically, they require affine equivariance)....
[...]