valida_convexhull.py

# -*- coding: utf-8 -*-

from PIL import Image, ImageFilter, ImageOps
from scipy import ndimage, fftpack
import pylab as p
import numpy as n
import imtools
import pickle as pk
from skimage.morphology import watershed, is_local_maximum, medial_axis, disk, convex_hull_image
from skimage.segmentation import *
from skimage.util import img_as_float
from skimage.filter import canny, threshold_otsu, rank
from scipy.misc import imsave
from sklearn.feature_extraction import image
from sklearn.cluster import spectral_clustering
import matplotlib.pyplot as plt

NUM_PINTURAS = 1

# artistas = ['Pollock', 'Picasso',
#             'Miro', 'Gogh', 'Kandinsky', 'Matisse',
#             'Poussin', 'Caravaggio', 'Hals', 'Rembrandt', 'Velazquez',
#             'Vermeer']

for teste in xrange(1,11):
    # abre a imagem
    im_rgb = Image.open('valida_convexhull/%s.jpg' % teste)

    im_cinza = im_rgb.convert('L')

    # equaliza
    #im_eq = ImageOps.equalize(im_cinza)
    
    ### FILTRAGEM
    
    # filtro media
    #im_media = Image.fromarray(ndimage.median_filter(im_eq, size=3))
    
    im_binaria = 255.*(n.array(im_cinza)>128)

    im_binaria[im_binaria == 255] = 1
    
    # calculamos o convex hull
    convex = convex_hull_image(im_binaria)
    
    # area original do segmento
    Aor = n.sum(im_binaria)
    # area do convex hull
    Ac = n.sum(convex)
    print Aor, Ac, float(Aor) / Ac
    
    #plt.clear()
    plt.imshow(convex, cmap=plt.cm.gray, interpolation='nearest')
    plt.savefig('valida_convexhull/%s_convex.png' % teste)
    plt.imshow(im_binaria, cmap=plt.cm.gray, interpolation='nearest')
    plt.savefig('valida_convexhull/%s_orig.png' % teste)
    #plt.show()