Parallel Computation of Component Trees on Distributed Memory Machines

Abstract

Component trees are region-based representations that encode the inclusion relationship of the threshold sets of an image. These representations are one of the most promising strategies for the analysis and the interpretation of spatial information of complex scenes as they allow the simple and efficient implementation of connected filters. This work proposes a new efficient hybrid algorithm for the parallel computation of two particular component trees-the max-and min-tree-in shared and distributed memory environments. For the node-local computation a modified version of the flooding-based algorithm of Salembier is employed. A novel tuple-based merging scheme allows to merge the acquired partial images into a globally correct view. Using the proposed approach a speed-up of up to 44.88 using 128 processing cores on eight-bit gray-scale images could be achieved. This is more than a five-fold increase over the state-of-the-art shared-memory algorithm, while also requiring only one-thirty-second of the memory.

Distributed Connected Component Filtering and Analysis in 2D and 3D Tera-Scale Data Sets

“…Parallelization strategies using distributed memory machines are more complex because one needs to make sure that the communication between the individual nodes does not become prohibitive during the tree construction. Two different approaches have emerged in the past few years: the Distributed Component Forests (DCF) [11], [12] and the distributed computation of the Component Tree [13]. Both use a similar divide-and-conquer approach, where the image is first divided in N p tiles, assigned to N p individual nodes.…”

“…In the approach of Götz et al [13], the entire Component Tree is computed and stored in a distributed manner. Hence, each process contains a subset of the entire component tree, such that a tree node in process p might point to a parent in an other process.…”

“…Gazagnes & Wilkinson [12] extended these techniques to 16 bit per pixel (bpp) data sets, and implemented multiscale operators within the DCF framework. Recently, Götz et al proposed a novel approach to compute component trees using both shared-memory and distributed memory techniques [13]. However, the latter does not include any attribute computation, filtering or multi-scale operations, limiting its application.…”

Distributed Component Forests in 2-D: Hierarchical Image Representations Suitable for Tera-Scale Images

“…Quite apart from that limitation, communicating entire component trees becomes prohibitive. In a recent work G€ otz et al 8 implemented a method to perform a parallel computation of component trees on distributed memory machines. Theȳ rst compute the local component trees of each tile, and then use a di®erent approach based on a speci¯c data structure, the tupples, to perform a parallel correction of the parents in each local component tree.…”

Automated Analysis of Remotely Sensed Images Using the Unicore Workflow Management System

“…In the first step, Morphological Attribute Profiles (APs) compute features that characterize the spatial information within a given image [7,8]. The APs result from a sequential application of attribute filters based on component trees which can be computed a the shared-and distributed-memory hybrid algorithm proposed by some of the authors [9]. This implementation outperforms traditional serial algorithms whose performances are strongly affected by the size and the quantization of the data.…”