#### DMCA

## Automatic Image Segmentation by Dynamic Region Merging (2011)

Venue: | IEEE Trans. On Image Processing |

Citations: | 7 - 0 self |

### Citations

4591 | F.: ‘A computational approach to edge detection
- Canny
- 1986
(Show Context)
Citation Context ...n errors in the matching process, which is a reasonable consideration in the correspondence of ground-truth and segmentation results. In Table 4, we compared the F-measures of the Canny edge detector =-=[5]-=-, the graph-based method [18], the mean-shift method [32] and the DRM algorithm. The best parameter is chosen for each of these methods, and we have the F-measures as 0.62 for Canny edge detector, 0.6... |

3826 | Dynamic Programming - Bellman - 1957 |

3722 | Normalized cuts and image segmentation
- Shi, Malik
- 1997
(Show Context)
Citation Context ..., the region homogeneity is used as an important segmentation criterion. Many cut criteria in graph theory have been studied for this purpose. The most widely used cut criteria include normalized cut =-=[8]-=-, ratio cut [9], minimum cut [10] and so on. The aim of these algorithms is to produce a desirable segmentation by achieving global optimization of some cost functions. However, these cost functions o... |

2090 | Fast approximate energy minimization via graph cuts - Boykov, Veksler, et al. - 2001 |

1289 | An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision - Boykov, Kolmogorov - 2001 |

1004 |
Watersheds in digital spaces: an efficient algorithm based on immersion simulations
- Vincent, Soille
- 1991
(Show Context)
Citation Context ..., and using regions for merging can improve a lot the computational efficiency. For simplicity and in order to validate the effectiveness of the proposed DRM algorithm, we use the watershed algorithm =-=[31]-=- (with some modification) to obtain the initially over-segmented regions (please refer to Section VI-A for more information), yet using a more sophisticated initial segmentation algorithm (e.g. mean-s... |

989 |
Computer Vision: A Modern Approach
- Forsyth, Ponce
- 2002
(Show Context)
Citation Context ...o partition an image into a certain number of pieces which have coherent features (color, texture, etc.) and in the meanwhile to group the meaningful pieces together for the convenience of perceiving =-=[1]-=-. In many practical applications, as a large number of images are needed to be handled, human interactions involved in the segmentation process should be as less as possible. This makes automatic imag... |

933 | A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics
- Martin, Fowlkes, et al.
- 2001
(Show Context)
Citation Context ...(Pb) images (Fig. 10(e)), where the darker pixels indicate more people marked them as a boundary. For more information about the generation of human labeled ground truth segmentation, please refer to =-=[37]-=-. Let’s then discuss about the quantitative measure of segmentation quality. This task might be as hard as segmentation itself. In the past decades, many researchers have been studying the supervised ... |

816 |
Sequential Analysis
- Wald
- 1947
(Show Context)
Citation Context ... More specifically, the extent of consistency tells whether or not the tested data belong to the same group. It is measured by two hypotheses according to the Sequential Probability Ratio Test (SPRT) =-=[28]-=-: a null hypothesis H0 “the tested data are consistent” and an alternative hypothesis H1 “the tested data are inconsistent”. The proposed similarity measure describes the affinity between two neighbor... |

711 | Principles of Gestalt Psychology - Koffka - 1935 |

614 | Learning to detect natural image boundaries using local brightness, color, and texture cues
- Martin, Fowlkes, et al.
- 2004
(Show Context)
Citation Context ...entation methods, a group of human segmented images for each test sample are provided in the Berkeley Segmentation Dataset (BSDS), and a boundary-based evaluation method was proposed by Martin et al. =-=[38]-=-. In [38], the precision-recall framework is applied in conjunction with the human-marked boundaries. The precision-recall is a well-accepted measure in image segmentation, which considers two aspects... |

353 | Using dynamic programming for solving variational problems in vision - Amini, Weymouth, et al. - 1990 |

349 | An optimal graph theoretic approach to data clustering: Theory and application to image segmentation
- Wu, Leahy
- 1993
(Show Context)
Citation Context ... as an important segmentation criterion. Many cut criteria in graph theory have been studied for this purpose. The most widely used cut criteria include normalized cut [8], ratio cut [9], minimum cut =-=[10]-=- and so on. The aim of these algorithms is to produce a desirable segmentation by achieving global optimization of some cost functions. However, these cost functions only provide a characterization of... |

248 |
Use of watersheds in contour detection
- Beucher, Lantuéjoul
- 1979
(Show Context)
Citation Context ...rging address the optimization of some global energy terms, such as the number of labels [43] and the area of regions [44]. As a good representation of morphological segmentation, watershed transform =-=[45]-=- can also be classified as region-based segmentation methods. The intuitive idea comes from geography, where watersheds are the dividing lines of different catchment basins. The major drawback of wate... |

168 | Associative hierarchical crfs for object class image segmentation
- Ladicky, Russell, et al.
- 2009
(Show Context)
Citation Context ...hould be as less as possible. This makes automatic image segmentation techniques more appealing. Moreover, the success of many high-level segmentation techniques (e.g. class-based object segmentation =-=[2, 3]-=-) also demands sophisticated automatic segmentation techniques. Dating back over decades, there is a large amount of literature on automatic image segmentation. For example, the edge detection algorit... |

140 | Hybrid image segmentation using watersheds and fast region merging - Haris, Efstratiadis, et al. - 1998 |

136 | From contours to regions: an empirical evaluation
- Arbelaez, Maire, et al.
- 2009
(Show Context)
Citation Context ... (with mean-shift as initial segmentation) for the proposed DRM method. Table 4. The best F-measures by the competing methods on the BSDS dataset. The average F-measure for the human subjects is 0.79 =-=[40]-=-, which represents the human performance for the segmentation task. The DRM algorithm is tested when initialized by watershed and mean-shift respectively. Graph-based Canny Mean-shift DRM (initialized... |

83 | Image segmentation with ratio cut
- Wang, Siskind
(Show Context)
Citation Context ...mogeneity is used as an important segmentation criterion. Many cut criteria in graph theory have been studied for this purpose. The most widely used cut criteria include normalized cut [8], ratio cut =-=[9]-=-, minimum cut [10] and so on. The aim of these algorithms is to produce a desirable segmentation by achieving global optimization of some cost functions. However, these cost functions only provide a c... |

68 | A hierarchical approach to color image segmentation using homogeneity - Cheng, Sun - 2000 |

66 |
Spatio-temporal segmentation based on region merging
- Moscheni, Bhattacharjee, et al.
- 1998
(Show Context)
Citation Context ...neity criterion is satisfied. In previous works, there are region merging algorithms based on statistical properties [16, 22-23, 50-51], graph properties [17-19, 24-25] and spatio-temporal similarity =-=[20]-=-. Although the segmentation is obtained by making local decisions, some techniques [16-21] have shown satisfying results with efficient implementation. Most region merging algorithms do not have some ... |

62 | Theories of Visual perception - Gordon |

47 | On Nearest-Neighbor Graphs
- Eppstein, Paterson, et al.
- 1997
(Show Context)
Citation Context ...rd its most similar neighbor. For a sequence of graph nodes, if the starting and ending nodes coincide, we call it a cycle (see Fig. 4). It is easy to verify that the NNG has the following properties =-=[36]-=-: (1) Along any directed edge in NNG, the weights are non-increasing. (2) The maximum length of a cycle is two. (3) NNG contains at least one cycle. (4) The maximum number of cycles is ⎣||V||/2⎦. From... |

40 | hierarchical segmentation and waterfall algorithm - Watershed - 1994 |

38 | Interactive image segmentation by maximal similarity based region merging - Ning, Zhang, et al. - 2010 |

35 | Canny edge detection enhancement by scale multiplication - Paul, Zhang, et al. - 2005 |

32 | Superpixel lattices - Moore, Prince, et al. - 2008 |

31 | Regions adjacency graph applied to color image segmentation
- Trémeau, Colantoni
(Show Context)
Citation Context ...e proposed predicate is based on measuring the dissimilarity between pixels along the boundary of two regions. For the convenience of expression, we use the definition of region adjacency graph (RAG) =-=[30]-=- to represent an image. Let G = (V, E) be an undirected graph, where vi∈V is a set of nodes corresponding to image elements (e.g. super-pixels or regions). E is a set of edges connecting the pairs of ... |

26 | Region merging techniques using information theory statistical measures
- Calderero, Marques
- 2010
(Show Context)
Citation Context ...order. Particularly, a homogeneity criterion is proposed based on the 3expectation of pixel colors inside a region, which naturally leads to a merging predicate. In the recent work of region merging =-=[22]-=-, a likelihood ratio test is applied as the measure of region similarity. The probability of adjacent regions and that of their merging are both computed. To minimize both probabilities of error, the ... |

24 | S.: Fast implementation of waterfall based on graphs - Marcotegui, Beucher |

22 |
Shift: A Robust Approach Toward Feature Space Analysis
- Commaniciu, “Mean
- 2002
(Show Context)
Citation Context ...h some modification) to obtain the initially over-segmented regions (please refer to Section VI-A for more information), yet using a more sophisticated initial segmentation algorithm (e.g. mean-shift =-=[32]-=-) may lead better final segmentation results. Given an over-segmented image, there are many regions to be merged for a meaningful segmentation. By taking the region merging as a labeling problem, the ... |

18 | Unsupervised multistage image classification using hierarchical clustering with a Bayesian similarity measure - Lee, Crawford - 2005 |

14 | General region merging approaches based on information theory statistical measures - Calderero, Marqués |

11 | A region merging prior for variational level set image segmentation
- Ayed, I, et al.
- 2008
(Show Context)
Citation Context ... have some desirable global properties, even though some recent works in region merging address the optimization of some global energy terms, such as the number of labels [43] and the area of regions =-=[44]-=-. As a good representation of morphological segmentation, watershed transform [45] can also be classified as region-based segmentation methods. The intuitive idea comes from geography, where watershed... |

10 | J.: Lattice cut - constructing superpixels using layer constraints - MOORE, PRINCE, et al. - 2010 |

6 |
Efficient graph-based image segmentation,”Int
- Felzemszwalb, Huttenlocher
- 2004
(Show Context)
Citation Context ...M algorithm can produce a segmentation S which is neither over-merged nor under-merged according to the proposed predicate P. Similar to the definitions of over-segmentation and under-segmentation in =-=[18]-=-, we define the concepts of over-merged segmentation and under-merged segmentation as below. Definition 1. Under-merged segmentation. A segmentation S is under-merged if it contains some pair of regio... |

5 | Simultaneous object classification and segmentation with high-order multiple shape models,‖ IEEETrans
- Lecumberry, Pardo, et al.
- 2010
(Show Context)
Citation Context ...hould be as less as possible. This makes automatic image segmentation techniques more appealing. Moreover, the success of many high-level segmentation techniques (e.g. class-based object segmentation =-=[2, 3]-=-) also demands sophisticated automatic segmentation techniques. Dating back over decades, there is a large amount of literature on automatic image segmentation. For example, the edge detection algorit... |

5 | Iterated graph cuts for image segmentation - Peng, Zhang, et al. |

5 | Homogeneity testing for unlabeled data: A performance evaluation,‖ CVGIP: Graph - Wu - 1993 |

5 |
Hossam Isack, and Yuri Boykov. Fast Approximate Energy Minization with Label Costs. International journal of computer vision
- Delong, Osokin
(Show Context)
Citation Context ...ion merging algorithms do not have some desirable global properties, even though some recent works in region merging address the optimization of some global energy terms, such as the number of labels =-=[43]-=- and the area of regions [44]. As a good representation of morphological segmentation, watershed transform [45] can also be classified as region-based segmentation methods. The intuitive idea comes fr... |

4 | Accurate boundary localization using dynamic programming on snake - Mishra, Fieguth, et al. - 2008 |

4 | Bayesian region merging probability for parametric image models - LaValle, Hutchinson - 1993 |

3 | Scene shape priors for superpxiel segmentation - Moore, Prince, et al. |

3 | Statistical multiscale image segmentation via alpha-stable modeling - Wan, Canagarajah, et al. |

2 | Fast image segmentation using region merging with a k-nearest neighbor graph - Liu, Guo, et al. |

1 | et al, “Edge detection by scale multiplication in wavelet domain - Zhang, Paul - 2002 |

1 | Segmentation of laparoscopic images: integrating graph-based segmentation and multistage region merging. The 2nd Canadian Conference on Computer and Robot Vision - Shu, Bilodeau, et al. - 2005 |

1 | Sequential algorithms for cell segmentation: maximum efficiency - Fernand - 1986 |

1 |
Hierarchies of partitions and morphological segmentation
- Fernand
- 2001
(Show Context)
Citation Context ...tation of the image. To overcome this problem, one solution is “flooding” from the selected markers [46-48] such that only the most important regional minima are saved for the segmentation. The other =-=[49]-=- is based on a hierarchical process, where the catchment basins of the watershed image are merged until they belong to almost homogeneous regions. In this paper, we implement the segmentation algorith... |