#### DMCA

## A Proposal for a Parallel Watershed Transform Algorithm for Real-Time Segmentation

### Citations

1015 |
Watershed in digital spaces: An efficient algorithm based on immersion simulations
- Vincent, Soille
- 1993
(Show Context)
Citation Context ... use of this algorithm. Lastly, Sec. 6 discuss the conclusions and perspectives of use along with future works. 2. Watershed Transform Many definitions for the watershed transform exist in literature =-=[3, 6, 7, 12, 15, 24]-=- that take different approaches on the problem, such as defining connected components via influence zones, shortest-path forests with a custom distance function and locally, by making paths of steepes... |

242 |
Scalable parallel programming with cuda. Queue 2008;6(2):40–53. doi:http://www.doi.acm.org/10.1145/ 1365490.1365500. 94 R. Mehra et al.
- Nickolls, Buck, et al.
- 2012
(Show Context)
Citation Context ...sation. Fig. 2 presents the programming platform for CUDA and C/C++ proposed by Halfhill [10]. As a complex topic out of the scope of this paper, the reader is referred to the work of Nickolls et al. =-=[17]-=- for a detailed view on CUDA programming and modelling. 4. Proposed Algorithm In this section our proposed algorithm is exposed and explained. Firstly, the motivations and inspirations are presented. ... |

226 | The watershed transform: Definitions, algorithms and parallelization strategies
- Roerdink, Meijster
- 2001
(Show Context)
Citation Context ...lly, the focus was on typical image processing systems, where the segmentation represented a time consuming operation [2,13]. Roerdink and Meijster extensively surveyed the literature on this problem =-=[21]-=-. More recently, the evolution of sequential algorithms along with hardware minimised this problem. However, for real-time applications (e.g. surveillance and navigation) the sequential algorithms are... |

125 |
Topographic distance and watershed lines
- Meyer
- 1994
(Show Context)
Citation Context ... use of this algorithm. Lastly, Sec. 6 discuss the conclusions and perspectives of use along with future works. 2. Watershed Transform Many definitions for the watershed transform exist in literature =-=[3, 6, 7, 12, 15, 24]-=- that take different approaches on the problem, such as defining connected components via influence zones, shortest-path forests with a custom distance function and locally, by making paths of steepes... |

95 | The image foresting transform theory, algorithms, and applications
- Falcão, Stolfi, et al.
(Show Context)
Citation Context ... use of this algorithm. Lastly, Sec. 6 discuss the conclusions and perspectives of use along with future works. 2. Watershed Transform Many definitions for the watershed transform exist in literature =-=[3, 6, 7, 12, 15, 24]-=- that take different approaches on the problem, such as defining connected components via influence zones, shortest-path forests with a custom distance function and locally, by making paths of steepes... |

62 | The Ordered Queue and the Optimality of the Watershed Approaches.
- Lotufo, Falcao
- 2000
(Show Context)
Citation Context |

47 | M.: Watershed cuts: Minimum spanning forests and the drop of water principle
- Cousty, Bertrand, et al.
- 2009
(Show Context)
Citation Context |

32 |
An efficient watershed algorithm based on connected components,
- Bieniek, Moga
- 2000
(Show Context)
Citation Context ...terature. 2.1. Sequential Watershed Algorithms The recent fastest sequential watershed transform algorithms are the result of the evolution of the arrowing technique for watershed of Bieniek and Moga =-=[3, 4]-=- and the union-find one of Meijster and Roerdink [14]. Several algorithms based on this preliminary works have been proposed, using variations of the previous procedures, achieving considerable speedu... |

19 |
Implementation of a distributed watershed algorithm
- Moga, Viero, et al.
- 1994
(Show Context)
Citation Context ...hitecture present a new field for the development of watershed algorithms. The work on parallel watershed algorithms was introduced by Moga et al. for reduction of time of processing for large images =-=[16]-=-. With advances on CPU speed, the focus changed more recently to real-time image segmentation, either on clusters or FPGAs [8, 23]. In this sense, our work, to the best of our knowledge, is innovative... |

15 | A disjoint set algorithm for the watershed transform,” in Procs. of Eusipco : European signal processing conference
- Meijster, Roerdink
- 1998
(Show Context)
Citation Context ...n’s [22], with the point-out and point-in strategy for storing the implicit lower-complete graph of the image. Also, the Union-Find strategy used by Meijster and Roerdink is used for minima labelling =-=[14]-=-. On the parallel field, the strategy of Galilée et al. [8] for local decisions through message passing is used, although in a different way, without explicit messages. This paper is organised as fol... |

12 | Comparison between immersion-based and toboggan-based watershed image segmentation
- Lin, Tsai, et al.
- 2006
(Show Context)
Citation Context ...ijster and Roerdink [14]. Several algorithms based on this preliminary works have been proposed, using variations of the previous procedures, achieving considerable speedups without loss of precision =-=[6, 11, 20, 22]-=-. These algorithms are all based on evaluating the neighbourhood, intuitively, to identify the direction of sliding of a drop of water on a surface until it reaches a minimum, and label the regions wh... |

11 | A connected component approach to the watershed segmentation
- Bieniek, Moga
- 1998
(Show Context)
Citation Context |

9 | A parallel watershed algorithm
- Bieniek, Burkhardt, et al.
- 1997
(Show Context)
Citation Context ...y demanding task, early studies have been developed on parallel algorithms. Initially, the focus was on typical image processing systems, where the segmentation represented a time consuming operation =-=[2,13]-=-. Roerdink and Meijster extensively surveyed the literature on this problem [21]. More recently, the evolution of sequential algorithms along with hardware minimised this problem. However, for real-ti... |

9 | A Proposal for the Implementation of a Parallel Watershed Algorithm
- Meijster, Roerdink
- 1995
(Show Context)
Citation Context ...y demanding task, early studies have been developed on parallel algorithms. Initially, the focus was on typical image processing systems, where the segmentation represented a time consuming operation =-=[2,13]-=-. Roerdink and Meijster extensively surveyed the literature on this problem [21]. More recently, the evolution of sequential algorithms along with hardware minimised this problem. However, for real-ti... |

9 |
An improved watershed algorithm based on efficient computation of shortest paths,”
- Osma-Ruiz, Godino-Llorente, et al.
- 2007
(Show Context)
Citation Context ...ijster and Roerdink [14]. Several algorithms based on this preliminary works have been proposed, using variations of the previous procedures, achieving considerable speedups without loss of precision =-=[6, 11, 20, 22]-=-. These algorithms are all based on evaluating the neighbourhood, intuitively, to identify the direction of sliding of a drop of water on a surface until it reaches a minimum, and label the regions wh... |

5 |
P.Y.: Parallel asynchronous watershed algorithm—architecture
- Galilee, Mamalet, et al.
- 2007
(Show Context)
Citation Context ...uced by Moga et al. for reduction of time of processing for large images [16]. With advances on CPU speed, the focus changed more recently to real-time image segmentation, either on clusters or FPGAs =-=[8, 23]-=-. In this sense, our work, to the best of our knowledge, is innovative as we propose an algorithm for implementation on off-the-shelf video cards. This work has been inspired by several other algorith... |

4 |
A fast watershed algorithm based on chain code and its application in image segmentation.
- Sun, Yang, et al.
- 2005
(Show Context)
Citation Context ...e propose an algorithm for implementation on off-the-shelf video cards. This work has been inspired by several other algorithms. Among the sequential ones, the most influential is Sun, Yang and Ren’s =-=[22]-=-, with the point-out and point-in strategy for storing the implicit lower-complete graph of the image. Also, the Union-Find strategy used by Meijster and Roerdink is used for minima labelling [14]. On... |

3 |
Programming Guide 2.1
- CUDA
- 2008
(Show Context)
Citation Context ...omparing the computational power of an Intel CPU, measured in peak GFlops, to the NVIDIA graphics cards, where the most modern GPU architecture delivers performance up to 6 times higher than the CPUs =-=[18]-=-. To explore the potential of the GPU, a different paradigm of programming should be used, called programming flow. Data are packaged in streams and the arithmetic calculations are kernels operating o... |

2 | Real-time image segmentation based on a parallel and pipelined watershed algorithm - Trieu, Maruyama - 2007 |

1 |
Gpu based image segmentation livewire algorithm implementation
- Baggio
- 2007
(Show Context)
Citation Context ...ackaged in streams and the arithmetic calculations are kernels operating on them. Excerpts of programming that have enormous arithmetic rates can be shared in order to use the most of the GPU. Baggio =-=[1]-=- characterises the structure of algorithms as follows: (1) the parallel sections of the program are identified and implemented with a kernel, which is a share of the GPU to process arrays of data in p... |

1 |
Gpu programming – syllabus. http://www.ccs.neu.edu/course/ csu610/#syllabus
- Cooperman, Kaeli
- 2009
(Show Context)
Citation Context ...ing called SIMD (Single Instruction Multiple Data). For this card, execution times of up to 100 times faster than the CPU time in classical programs as a multiplication of matrices have been obtained =-=[5]-=-. A study by NVIDIA presented a chart comparing the computational power of an Intel CPU, measured in peak GFlops, to the NVIDIA graphics cards, where the most modern GPU architecture delivers performa... |

1 |
Parallel processing with cuda: Nvidia’s high-performace computing platform uses massive multithreading
- Halfhill
(Show Context)
Citation Context ... extended from C. Its main abstraction is based on the hierarchy of thread groups, memory sharing and synchronisation. Fig. 2 presents the programming platform for CUDA and C/C++ proposed by Halfhill =-=[10]-=-. As a complex topic out of the scope of this paper, the reader is referred to the work of Nickolls et al. [17] for a detailed view on CUDA programming and modelling. 4. Proposed Algorithm In this sec... |

1 |
technical training. volume i: Introduction to cuda programming
- Cuda
(Show Context)
Citation Context ...ng and obtain the best performance. The image block size must be a multiple of the warp size, which is characterised as a minimum group of 64 threads, processed as SIMD by CUDA many-core architecture =-=[19]-=-. Empirically, it is suggested the use of 256 threads as a good choice, balancing between memory latency, registers and threads. Another parameter is the memory access. For this development, the textu... |