img_prepare.py

import numpy as np
from skimage.color import rgb2gray
from skimage.filters import gaussian
import scipy.ndimage as ndi
import random
import scipy
import PIL

def img_prepare(img):
    r = random.random()
	
    if r<=0.3:
        img = img.quantize()
        img = img.convert("RGB")
    
    r = random.random()
    if r>=0.5:
        return img

    px = np.asarray(img, dtype = np.int32)
	
    gs = rgb2gray(px/255.0)
    
    markers = np.zeros_like(gs)
    markers[gs < 0.1] = 1
    markers[gs > 0.6] = 2

    from skimage.filters import sobel
    elevation_map = sobel(gs)

    from skimage.morphology import watershed
    segmentation = watershed(elevation_map, markers)

    color = (int(random.random()*255),int(random.random()*255),int(random.random()*255))
    if r<0.2:
        px[segmentation>1,0:3]=color
    elif r<0.5: 
        px[segmentation==1,0:3]=color

    return PIL.Image.fromarray( np.asarray( np.clip(px,0,255), dtype=np.int8), "RGB" )