Cellular automaton for ultra-fast watershed transform on GPU

Abstract

In this paper we describe a cellular automaton (CA) used to perform the watershed transform in N-D images. Our method is based on image integration via the Ford-Bellman shortest paths algorithm. Due to the local nature of CA algorithms we show that they are designed to run on massively parallel processors and therefore, be efficiently implemented on low cost consumer graphical processing units (GPUs).

Efficient GPU Asynchronous Implementation of a Watershed Algorithm Based on Cellular Automata

“…The experiments are carried out over volume datasets obtaining maximum speedups of only 7x on a Nvidia GTX295 when comparing to the sequential implementation of the algorithm, even when large volume datasets of up to 600 × 600 × 600 are considered. The algorithm presented in [11] is inspired by the drop of water paradigm and depth-search approaches and compares its results to a synchronous algorithm for the watershed based on a CA [10] outperforming it. Given that the experiments in [11] are performed in a older GPU, we have executed them on the GTX580 obtaining similar speedup results to the obtained with our asynchronous implementation of the CA-watershed described in this paper.…”

“…The Computed Unified Device Architecture (CUDA) developed by NVIDIA [8], based on a data parallel programming model, provides support for general-purpose computing on graphics hardware [9]. In recent years a number of parallel implementations of watershed algorithms in GPUs have been published [10]- [12].…”

“…Watershed implementations based on CA were proposed for an array processor [7], and using shaders in the graphic pipeline of the GPU [10]. Both automata are synchronous because they update the entire image at each step of the evolution of the automaton.…”

Medical image processing on the GPU – Past, present and future

“…Pixels that are classified as local minima are then collected into regions, an operation which may require communication between different thread blocks. Kauffmann and Piche (2008) instead used a cellular automaton (a collection of simple processing cells arranged on a regular lattice) approach to perform watershed segmentation. A volume of the size 512 × 512 × 512 voxels could be processed in about 17 seconds.…”

Seeded ND medical image segmentation by cellular automaton on GPU

“…In a previous published paper [32], we introduced our GPU-FBA based segmentation approach to compute the watershed transform on ND images. In the present work, we demonstrate that this approach can efficiently be used to perform multi-label seeded organ segmentation in ND images.…”