RegionGrowth - 2d or 3d region growing on gray scale images

features

• Supports one seed or many seeds as the input of RegionGrowth algorithm

• Supports 2d region growth: RegionGrowth.RegionGrow2D

  • 4-neighbor
  • 8-neighbor

• Supports 3d region growth: RegionGrowth.RegionGrow3D

  • 6-neighbor
  • 26-neighbor

• Supports adaptive thresholding method by taking (mean±3*std) of foreground pixel values as thresholds

• Supports constant lower and upper thresholds

install

python setup.py build_ext

It will generate "RegionGrowth.xxx.pyd" in "./build" path, which can be imported in python.

usage

import RegionGrowth
RegionGrowth.RegionGrow2D(img, masks, upperThreshold=255, lowerThreshold=0, neighborMode="8n").main(seeds, update=True)

• img: np.ndarray[np.uint8, ndim=2] or np.ndarray[np.uint8, ndim=3] (only writeable array is supported).
• masks: same shape and type with img. determining whether pixels are required to be checked. If there are some given regions that doesn't contain foreground pixels, you can set corresponding masks to indicate that; otherwise, use masks = np.ones_like(img).
• upperThreshold and lowerThreshold: constant thresholds
• neighborMode: RegionGrow2D supports "4n" or "8n"; RegionGrow3D supports "6n" or "26n";
• seeds: the coordinates of seeds [(y,x)]
• update: whether enable adaptive thresholding method by taking (mean±3*std) of foreground pixel values as thresholds.

demo

import RegionGrowth
import numpy as np
from PIL import Image


if __name__ == "__main__":
    img = Image.open("images/input.png").convert('L') # gray scale
    img = np.array(img)

    masks = np.ones_like(img)

    seed_img = Image.open("images/random.png").convert('L')
    seed_img = np.array(seed_img)
    coords = np.where(seed_img>0)

    seeds = np.stack(coords, axis=-1)

    out_img = RegionGrowth.RegionGrow2D(img, masks, upperThreshold=255, lowerThreshold=0, neighborMode="8n").main(seeds, update=True)
 
    out_img = Image.fromarray(np.asarray(out_img)*255)
    out_img.save("outputs/result_growth_pyx.png")

results

input image

Ground-truth

seeds

output

---

2020-04-10