Abstruct

Implemnetation of Graph Based Segmentation Algorithm with Union Find. Refer a paper below.

http://cs.brown.edu/~pff/papers/seg-ijcv.pdf

Usage

Refer a file GraphBasedSegmentTest.py

1. Import Image(PIL), numpy, SegmentationProcess.

   • from PIL import Image
   • import numpy as np
   • from SegmentationProcess import *

2. Load an image with PIL.

   • img = np.array(Image.open('pathOfImage.png'))

3. Init SegmentationProcessClass (GridGraph or NearestNeighborGraph)

   • ggs = GridGraphSegmentation(src_img, dst_img, top_n, tau_k)
   • nngs = NearestNeighborSegmentation(src_img, dst_img, top_n, tau_knn)

4. Call a "train" method to get a processed image.

   • ggs.train()
   • nngs.train()

Precautions

1. NearestNeighbor takes far more time than Grid Graph. Grid Graph method considers 8 Nearest pixel as edge. NearestNeighbor considers pixels the distance of which is less than n.

2. Parameter top n controls how many merged components are to be colored. Merged Components to be colored are selected with the descending order of number of cells contained in the component.

3. Parameter nn controls the distance in which pixels are considered to be edges. Cosidering the cell (r, c) and nn, cells (rr, cc) which meet the condition sqrt((r-rr)^2+(c-cc)^2) < nn are considered to be edges of the cell.

4. (Optional) You can specify segmentation tau parameter. Components can be more easily merged with larger tau value.