## Robust real-time segmentation of images and videos uisng a smoothing-spline snake-based algorithm (2005)

Venue: | IEEE Transactions on Image Processing |

Citations: | 22 - 5 self |

### BibTeX

@ARTICLE{Precioso05robustreal-time,

author = {Frederic Precioso and Michel Barlaud and Thierry Blu and Michael Unser},

title = {Robust real-time segmentation of images and videos uisng a smoothing-spline snake-based algorithm},

journal = {IEEE Transactions on Image Processing},

year = {2005},

volume = {14},

pages = {2005}

}

### OpenURL

### Abstract

Abstract—This paper deals with fast image and video segmentation using active contours. Region-based active contours using level sets are powerful techniques for video segmentation, but they suffer from large computational cost. A parametric active contour method based on B-Spline interpolation has been proposed in [26] to highly reduce the computational cost, but this method is sensitive to noise. Here, we choose to relax the rigid interpolation constraint in order to robustify our method in the presence of noise: by using smoothing splines, we trade a tunable amount of interpolation error for a smoother spline curve. We show by experiments on natural sequences that this new flexibility yields segmentation results of higher quality at no additional computational cost. Hence, real-time processing for moving objects segmentation is preserved. I.

### Citations

3402 | Snakes: active contour models - Kass, Witkin, et al. - 1998 |

1190 | Geodesic active contours
- Caselles, Kimmel, et al.
- 1997
(Show Context)
Citation Context ...ion framework by choosing a criterion which includes region and boundary functionals. Boundary functionals were first proposed by Kass et al. [21] and geodesic active contours by Caselles et al. [3], =-=[4]-=- for active contour segmentation. Region-based active contours were first introduced by Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information ... |

1174 |
Fronts propagating with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations
- Osher, Sethian
- 1988
(Show Context)
Citation Context ...oving objects. • is a constant. B. Implementation Region-based active contour evolution can be implemented in two different ways. (2) (3) (4) • Implicitly, based on the level-set approach [19], [20], =-=[23]-=-. Such a method provides an implicit management of topological changes and yields accurate results, but it suffers from a high computational cost. • Explicitly, using active parametric contours. Such ... |

913 | Active contours without edges
- Chan, Vese
- 2001
(Show Context)
Citation Context ...t introduced by Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. =-=[6]-=-, Zhu et al. [35], Paragios et al. [24], and Debreuve et al. [12] introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. [17], [20] address the segmentation ... |

681 | Region competition: unifying snakes, region growing, and bayes/mdl for multiband image segmentation
- Zhu, Yuille
- 1996
(Show Context)
Citation Context ...Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. =-=[35]-=-, Paragios et al. [24], and Debreuve et al. [12] introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. [17], [20] address the segmentation problem where fea... |

289 |
An Introduction to Splines for use in Computer Graphics and Geomteric Modeling
- Bartels, Beatty, et al.
- 1987
(Show Context)
Citation Context ...connected between segments. Here, we assume that there are such segments parametered by with the assumption and that we are given the sampling points . Each segment is expressed as a cubic polynomial =-=[1]-=- is a nonuniform B-spline function; the parameters of the model are the B-spline coefficients called control points. These coefficients can be specified by solving for the equations . The B-spline fun... |

257 | Geodesic active regions and level set methods for supervised texture segmentation
- Paragios, Deriche
- 2002
(Show Context)
Citation Context ...d Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et al. =-=[24]-=-, and Debreuve et al. [12] introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. [17], [20] address the segmentation problem where features of the region to... |

253 | De- terministic edge-preserving regularization in computed imaging
- Charbonnier, Blanc-Feraud, et al.
- 1997
(Show Context)
Citation Context ...ECIOSO et al.: ROBUST REAL-TIME SEGMENTATION OF IMAGES AND VIDEOS 917 Fig. 10. Segmentation through a sequence. (a) Frame 10. (b) Frame 20. (c) Frame 30. (d) Frame 40. estimation using motion vectors =-=[7]-=-. The motion field is evaluated by a classical block-matching algorithm between frames and [18]. and are two positive constants. The descriptors are region-independent. Thus, the local terms of the di... |

214 |
A geometric model for active contours in image processing
- Caselles, Catte, et al.
- 1993
(Show Context)
Citation Context ...mization framework by choosing a criterion which includes region and boundary functionals. Boundary functionals were first proposed by Kass et al. [21] and geodesic active contours by Caselles et al. =-=[3]-=-, [4] for active contour segmentation. Region-based active contours were first introduced by Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region informa... |

189 |
Smoothing by Spline Functions
- Reinsch
- 1967
(Show Context)
Citation Context ...ation method; namely, the smoothing spline method [28]. B. Cubic Spline Approximation A smoothing spline is an approximation curve controlled by a parameter trading interpolation error for smoothness =-=[29]-=-. It minimizes the following criterion: (12) where the s are the measured data points and s are points, on the curve, joining of polynomial pieces. The result is still a cubic spline, but it does not ... |

157 | Interpolation revisited
- Unser
- 2000
(Show Context)
Citation Context ...on of the criterion and computation of the velocity vector. In Section III, we propose a cubic B-spline implementation. Cubic B-splines preserve regularity and have excellent approximation properties =-=[32]-=- which means that, for a given accuracy, fewer samples are needed than with other parametric methods; moreover, fast algorithms are available for B-splines, which greatly reduces the computation cost.... |

142 |
Region-based strategies for active contour models
- Ronfard
(Show Context)
Citation Context ... were first proposed by Kass et al. [21] and geodesic active contours by Caselles et al. [3], [4] for active contour segmentation. Region-based active contours were first introduced by Ronfard et al. =-=[30]-=- and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et a... |

136 | B-spline signal processing: Part I-Theory
- Unser, Aldroubi, et al.
- 1993
(Show Context)
Citation Context ... fast algorithms are available for B-splines, which greatly reduces the computation cost. Unfortunately, interpolation methods are not robust to noise. This is why we propose to use smoothing splines =-=[33]-=- in the B-spline interpolation approach of [27]. These curves preserve the implementation advantages as the B-splines while softening the interpolation constraint. The relaxation of the interpolation ... |

116 |
Deformable boundary finding in medical images by integrating gradient and region information
- Chakraborty, Staib, et al.
- 1996
(Show Context)
Citation Context ...odesic active contours by Caselles et al. [3], [4] for active contour segmentation. Region-based active contours were first introduced by Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. =-=[5]-=- combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et al. [24], and Debreuve et al. [12] introduce re... |

96 | Using prior shapes in geometric active contours in a variational framework
- Chen, Tagare, et al.
- 2002
(Show Context)
Citation Context ...nal approach as opposed to previous work on shape prior, based on probabilistic methods [11]. The shape prior allows free form deformation [13] and is not restricted to a parametric deformation as in =-=[8]-=-. All these contour or region-based methods use a level-set approach which is accurate but time consuming. In this paper, we Manuscript received October 6, 2003; revised May 10, 2004. The associate ed... |

79 |
Statistical region snake-based segmentation adapted to different physical noise models
- Chesnaud, Réfrégier, et al.
- 1999
(Show Context)
Citation Context ...ontours were first introduced by Ronfard et al. [30] and Cohen et al. [10]. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. =-=[9]-=-, Chan et al. [6], Zhu et al. [35], Paragios et al. [24], and Debreuve et al. [12] introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. [17], [20] address ... |

71 | B-spline snakes: A flexible tool for parametric contour detection
- Brigger, Hoeg, et al.
(Show Context)
Citation Context ...ierry.blu@epfl.ch; michael.unser@epfl.ch). Digital Object Identifier 10.1109/TIP.2005.849307 1057-7149/$20.00 © 2005 IEEE propose a parametric active contour evolution based on a cubic spline contour =-=[2]-=-. In Section II, we present a survey of the region-based criterion, the derivation of the criterion and computation of the velocity vector. In Section III, we propose a cubic B-spline implementation. ... |

48 |
Nonparametric methods for image segmentation using information theory and curve evolution
- Kim, Fisher, et al.
- 2002
(Show Context)
Citation Context ...s. 2) Color Video Sequences: As detailed by Jehan et al. in [19], [20], the homogeneity in color images is related to the determinant of the covariance matrix for Gaussian distributions. Yezzi et al. =-=[22]-=- and Herbulot et al. [14] extended this framework to the more general case of Entropy descriptors without Gaussian distribution hypothesis. In these experiments, regions of interest are regions of hom... |

46 | Approximation with active B-spline curves and surfaces
- Pottmann, Leopoldseder, et al.
- 2002
(Show Context)
Citation Context ...arameter at the sampling points. Irregular sampling of is intuitively more pertinent, as regards active contour propagation, than uniform sampling. This is the option chosen by, e.g., Pottmann et al. =-=[25]-=- and Yang et al. [34], who propose to optimize the parameterization of the spline curve for approximating a target curve, as well as in other approaches based on arbitrary parameterizations (chord len... |

37 | Video object segmentation using eulerian region-based active contours
- Jehan-Besson, Barlaud, et al.
- 2001
(Show Context)
Citation Context ...Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et al. [24], and Debreuve et al. [12] introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. =-=[17]-=-, [20] address the segmentation problem where features of the region to be segmented are embedded in region functionals. In this framework, Gastaud et al. [13] propose a new approach introducing shape... |

34 | Surface reconstruction using active contour models
- Cohen, Bardinet, et al.
- 1993
(Show Context)
Citation Context ...y Kass et al. [21] and geodesic active contours by Caselles et al. [3], [4] for active contour segmentation. Region-based active contours were first introduced by Ronfard et al. [30] and Cohen et al. =-=[10]-=-. Chakraborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et al. [24], and Debreuve ... |

31 | Diffusionsnakes: Combining statistical shape knowledge and image information in a variational framework
- Cremers, Schnörr, et al.
- 2001
(Show Context)
Citation Context ... this framework, Gastaud et al. [13] propose a new approach introducing shape prior. This method uses a variational approach as opposed to previous work on shape prior, based on probabilistic methods =-=[11]-=-. The shape prior allows free form deformation [13] and is not restricted to a parametric deformation as in [8]. All these contour or region-based methods use a level-set approach which is accurate bu... |

31 |
Control point adjustment for B-spline curve approximation
- Yang, Wang, et al.
(Show Context)
Citation Context ...ing points. Irregular sampling of is intuitively more pertinent, as regards active contour propagation, than uniform sampling. This is the option chosen by, e.g., Pottmann et al. [25] and Yang et al. =-=[34]-=-, who propose to optimize the parameterization of the spline curve for approximating a target curve, as well as in other approaches based on arbitrary parameterizations (chord length, centripetal, Fol... |

23 |
Space time segmentation using level set active contours applied to myocardial gated SPECT
- Debreuve, Barlaud, et al.
- 2001
(Show Context)
Citation Context ...aborty et al. [5] combined both boundary and region information for medical images segmentation. Then, Chesnaud et al. [9], Chan et al. [6], Zhu et al. [35], Paragios et al. [24], and Debreuve et al. =-=[12]-=- introduce region-based statistic descriptors for image or video segmentation. Jehan-Besson et al. [17], [20] address the segmentation problem where features of the region to be segmented are embedded... |

17 | Combining shape prior and statistical features for active contour segmentation
- Gastaud, Barlaud, et al.
(Show Context)
Citation Context ... video segmentation. Jehan-Besson et al. [17], [20] address the segmentation problem where features of the region to be segmented are embedded in region functionals. In this framework, Gastaud et al. =-=[13]-=- propose a new approach introducing shape prior. This method uses a variational approach as opposed to previous work on shape prior, based on probabilistic methods [11]. The shape prior allows free fo... |

15 | Smoothing B-spline active contour for fast and robust image and video segmentation
- Precioso, Barlaud, et al.
- 2003
(Show Context)
Citation Context ... based on these B-spline curves [26], interpolation is not robust enough in the presence of noise. Thus, we propose to use a less constrained approximation method; namely, the smoothing spline method =-=[28]-=-. B. Cubic Spline Approximation A smoothing spline is an approximation curve controlled by a parameter trading interpolation error for smoothness [29]. It minimizes the following criterion: (12) where... |

14 |
2004a, ‘Shape gradient for image segmentation using information theory
- Herbulot, Jehan-Besson, et al.
(Show Context)
Citation Context ...es: As detailed by Jehan et al. in [19], [20], the homogeneity in color images is related to the determinant of the covariance matrix for Gaussian distributions. Yezzi et al. [22] and Herbulot et al. =-=[14]-=- extended this framework to the more general case of Entropy descriptors without Gaussian distribution hypothesis. In these experiments, regions of interest are regions of homogeneous color, like the ... |

13 |
Spline functions and the problem of graduation. Proc. of the National Academy of Sciences
- Schoenberg
- 1964
(Show Context)
Citation Context ... and the curvature, involved in the velocity equation, can be computed exactly at every sampling point. In addition, such curves minimize the following criterion: (11) under interpolatory constraints =-=[31]-=-. Here, is a parametric description of the curve and the second derivative of w.r.t. ; this functional is actually very close to the (squared) curvature when the parameter is close to the curvilinear ... |

11 | An exact method for computing the area moments of wavelet and spline curves
- Jacob, Blu, et al.
(Show Context)
Citation Context ...n. To improve the computation time of our algorithm with such descriptors based on area moments (mean of the intensity, variance, etc.), a perspective would be to implement the method of Jacob et al. =-=[15]-=- for an exact computation of the area moments of spline curves. 2) Color Video Sequences: As detailed by Jehan et al. in [19], [20], the homogeneity in color images is related to the determinant of th... |

9 | B-Spline active contours for fast video segmentation
- Precioso, Barlaud
- 2001
(Show Context)
Citation Context ...using level sets are powerful techniques for video segmentation, but they suffer from large computational cost. A parametric active contour method based on B-Spline interpolation has been proposed in =-=[26]-=- to highly reduce the computational cost, but this method is sensitive to noise. Here, we choose to relax the rigid interpolation constraint in order to robustify our method in the presence of noise: ... |

2 |
unifying approach and interface for spline-based snakes
- “A
- 2001
(Show Context)
Citation Context ...c description of the curve and the second derivative of w.r.t. ; this functional is actually very close to the (squared) curvature when the parameter is close to the curvilinear abscissa, as shown in =-=[16]-=-. Although we have obtained real-time accurate results with an implementation based on these B-spline curves [26], interpolation is not robust enough in the presence of noise. Thus, we propose to use ... |

2 |
3-step algorithm using region-based active contours for video objects detection
- “A
- 2002
(Show Context)
Citation Context ...rough a sequence. (a) Frame 10. (b) Frame 20. (c) Frame 30. (d) Frame 40. estimation using motion vectors [7]. The motion field is evaluated by a classical block-matching algorithm between frames and =-=[18]-=-. and are two positive constants. The descriptors are region-independent. Thus, the local terms of the differentiation (2) disappear and the velocity (4) reduces to (25) Since the descriptor is a temp... |

2 |
active contours with handling of topology changes for fast video segmentation
- “B-spline
- 2002
(Show Context)
Citation Context ...hich greatly reduces the computation cost. Unfortunately, interpolation methods are not robust to noise. This is why we propose to use smoothing splines [33] in the B-spline interpolation approach of =-=[27]-=-. These curves preserve the implementation advantages as the B-splines while softening the interpolation constraint. The relaxation of the interpolation condition is traded for an optimal increase of ... |