Research On Watershed-Parallel Segmentation Of High Resolution Remote Sensing Imagery

Image segmentation is the prerequisite and basis for image recognition and imageunderstanding, and as a region-based image segmentation method, the watershed segmentationmethod has the difficult in the process for segmentation and the significant multi-scale feature;with the remote sensing image data volume and complexity of the rapid growth, the speed ofimagery process and high accuracy is a new need. The thesis studies on image segmentation andparallel-processing general model for based on the analysis of domestic and foreign remotesensing imagery segmentation, and realizes the watershed parallel segmentation on the basis ofORFEO Toolbox (OTB).Main research results are as follows：1. To explore the feasibility of parallel processing for remote sensing data and concurrencybased on the processing and characteristics of the image. To propose the product-driven strategybased on task decomposition, the block-driven strategy based on the data decomposition, theimage processing chain-driven strategy based on functional decomposition and domain-drivenstrategy based on algorithm decomposition according to parallel algorithm design.2. The parallel-segmentation general model for remote sensing image is realized, using MPImaster-slave mode, according to the special nature of the image segmentation algorithm. Themodel uses task decomposition strategy to split the product development, functiondecomposition strategy to join the segmentation process, data decomposition strategy to blockdata, algorithm decomposition strategy to segmentation within a block. Experiments show thatthe model can effectively improve the speed and efficiency of image segmentation, so there iswidespread practical significance.3. To achieve the parallel watershed segmentation using parallel programming ideas andparallel processing strategies for remote sensing image based on open source library ORFEOToolbox (OTB) of distributed image processing algorithms.