This code tackle the problem of creating tiles for parallel segmentation. After defining tiles, any algorithm can run on them in an independent way, since the implementation is based on multiprogrammed techniques. The code define noncrisp borders between tiles, so it's not necessary a postprocessing of neighboring regions.
Anderson Soares, Thales Körting, Emiliano Castejon and Leila Fonseca Image Processing Division, National Institute of Space Research, INPE, São José dos Campos, Brazil
divSeg(img,filterOption,line_number,vchunk,hchunk,max_displacement,epsg)
img - input image
filterOption - 1 for standard magnitude approach, 2 for directional approach (recommended)
line_number - Nummber of lines to split the image (4 - divide in 16 tiles)
vchunk - Vertical chunk size
hchunk - horizontal chuck size
max_displacement - maximum displacement to the crop line
epsg - EPSG of image
weight - You can add a weight to a specific band to highlight some important feature.
divSeg('input.tif',8,2,20,10,100,32723,[1 1 1 1 1 2 2 2])
For geotiff files, it is necessary to provide the EPSG of image.
This project is licensed under the MIT License - see the LICENSE.md file for details
Divide And Segment - An Alternative For Parallel Segmentation. TS Korting, EF Castejon, LMG Fonseca - Lecture Notes in Computer Science 8192:504-515 · May 2013 PDF
Improvements of the divide and segment method for parallel image segmentation. AR Soares, TS Körting, LMG Fonseca - Revista Brasileira de Cartografia 68 (6) PDF