A Proposal for a ParallelWatershed Transform Algorithm for Real-Time Segmentation

André KorbesGiovani B. VitorJanito V. FerreiraRoberto de Alencar Lotufo

The watershed transform is widely used for image segmentationon computer vision applications. However, sequentialwatershed algorithms are not suitable for real-timeapplications, once they are one demanding part of severaltasks. This paper proposes a new parallel algorithm for thewatershed transform focused on real-time image segmentationusing off-the-shelf GPUs. In this sense, this algorithmaims for a speedup by mixing several techniques ofthe fastest procedures on both sequential and parallel fields.The algorithm has four major steps, processed on SIMD,with implementation details tuned for many-core architectures,without the use of explicit message passing schemes.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: