Jansen-MIDAS: a multi-level photomicrograph segmentation software based on isotropic undecimated wavelets

Alexandre Fioravante de Siqueira^{1, 2}, Flávio Camargo Cabrera², Wagner Massayuki Nakasuga^{1, 2}, Aylton Pagamisse³, Aldo Eloizo Job²

¹ Departamento de Raios Cósmicos e Cronologia, IFGW, University of Campinas

² Departamento de Física, Química e Biologia, FCT, UNESP - Univ Estadual Paulista

³ Departamento de Matemática e Computação, FCT, UNESP - Univ Estadual Paulista

Importing necessary packages.



In [1]:

    
from skimage.io import imread
from skimage.util import img_as_bool
from skimage.filters import (threshold_isodata, threshold_minimum, threshold_yen,
                             threshold_mean, threshold_triangle, threshold_li,
                             threshold_otsu)

from skimage.measure import compare_ssim
from skimage.transform import resize
from supmat_support import *

import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
import numpy as np
import warnings

Setting plot variables.



In [2]:

    
plt.rcParams['font.family'] = 'monospace'
plt.rcParams['font.size'] = 20
plt.rcParams['axes.labelsize'] = plt.rcParams['font.size']
plt.rcParams['axes.titlesize'] = 1.5*plt.rcParams['font.size']
plt.rcParams['legend.fontsize'] = 1.5*plt.rcParams['font.size']
plt.rcParams['xtick.labelsize'] = plt.rcParams['font.size']
plt.rcParams['ytick.labelsize'] = plt.rcParams['font.size']

warnings.filterwarnings('ignore')

Reading test images `test1.jpg` and `test2.jpg`.



In [3]:

    
img_test1 = imread('Figures/test1.jpg', as_grey=True)
img_test2 = imread('Figures/test2.jpg', as_grey=True)

fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(20, 10))
ax[0].imshow(img_test1, cmap='gray')
ax[0].set_xlabel('Test image 1.')

ax[1].imshow(img_test2, cmap='gray')
ax[1].set_xlabel('Test image 2.')
plt.show()

Reading the ground truth of the test images, `test1_GT.jpg` and `test2_GT.jpg`.



In [4]:

    
img_gt1 = img_as_bool(imread('Figures/test1_GT.jpg', as_grey=True))
img_gt2 = img_as_bool(imread('Figures/test2_GT.jpg', as_grey=True))

fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(20, 10))
ax[0].imshow(img_gt1, cmap='gray')
ax[0].set_xlabel('GT of test image 1.')

ax[1].imshow(img_gt2, cmap='gray')
ax[1].set_xlabel('GT of test image 2.')
plt.show()

Reading the MLSOS result of the test images, `test1_MLSOSL03L6.tif` and `test2_MLSOSL03L6.tif`.



In [5]:

    
img_mlsosbin1 = img_as_bool(imread('Figures/test1_MLSOSL03L6.tif', as_grey=True))
img_mlsosbin1 = img_as_bool(resize(img_mlsosbin1, output_shape=(884, 1024)))

img_mlsosbin2 = img_as_bool(imread('Figures/test2_MLSOSL03L6.tif', as_grey=True))
img_mlsosbin2 = img_as_bool(resize(img_mlsosbin2, output_shape=(576, 768)))

fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(20, 10))
ax[0].imshow(img_mlsosbin1, cmap='gray')
ax[0].set_xlabel('MLSOS result for test image 1 ($L_0 = 3$, $L = 6$).')

ax[1].imshow(img_mlsosbin2, cmap='gray')
ax[1].set_xlabel('MLSOS result for test image 2 ($L_0 = 3$, $L = 6$).')
plt.show()

Segmentation and comparison for test image 1.

Obtaining each threshold.



In [6]:

    
dataset_bin1 = [img_test1 > threshold_isodata(img_test1),
                img_test1 > threshold_minimum(img_test1),
                img_test1 > threshold_yen(img_test1),
                img_test1 > threshold_mean(img_test1),
                img_test1 > threshold_triangle(img_test1),
                img_test1 > threshold_li(img_test1),
                img_test1 > threshold_otsu(img_test1),
                img_mlsosbin1]

Obtaining SSIM for each threshold.



In [7]:

    
label_info1 = [' * ISODATA X GT: ', ' * Minimum X GT: ', ' * Yen X GT: ',
               ' * Mean X GT: ', ' * Triangle X GT: ', ' * Li X GT: ',
               ' * Otsu X GT: ', ' * MLSOS X GT: ']

print('SSIM - test1.jpg.')

for idx, img_bin1 in enumerate(dataset_bin1):
    print(label_info1[idx], compare_ssim(img_bin1, img_gt1))









    



SSIM - test1.jpg.
 * ISODATA X GT:  0.865700779506
 * Minimum X GT:  0.865700779506
 * Yen X GT:  0.815854134082
 * Mean X GT:  0.715361892034
 * Triangle X GT:  0.213501631907
 * Li X GT:  0.851034928043
 * Otsu X GT:  0.863225302782
 * MLSOS X GT:  0.785960559579

Visual comparison between the thresholds and the GT for the image 1.



In [8]:

    
dataset_vis1 = []

for idx, img_bin1 in enumerate(dataset_bin1):
    dataset_vis1.append(visual_comparison(img_bin1, img_gt1))



In [9]:

    
fig_comp1, ax_comp1 = plt.subplots(nrows=2, ncols=4, figsize=(22, 10))

ax_comp1[0, 0].imshow(dataset_vis1[0][0])
ax_comp1[0, 0].set_xlabel('ISODATA')
ax_comp1[0, 1].imshow(dataset_vis1[1][0])
ax_comp1[0, 1].set_xlabel('Minimum')
ax_comp1[0, 2].imshow(dataset_vis1[2][0])
ax_comp1[0, 2].set_xlabel('Yen')
ax_comp1[0, 3].imshow(dataset_vis1[3][0])
ax_comp1[0, 3].set_xlabel('Mean')
ax_comp1[1, 0].imshow(dataset_vis1[4][0])
ax_comp1[1, 0].set_xlabel('Triangle')
ax_comp1[1, 1].imshow(dataset_vis1[5][0])
ax_comp1[1, 1].set_xlabel('Li')
ax_comp1[1, 2].imshow(dataset_vis1[6][0])
ax_comp1[1, 2].set_xlabel('Otsu')
ax_comp1[1, 3].imshow(dataset_vis1[7][0])
ax_comp1[1, 3].set_xlabel('MLSOS')

plt.show()



In [10]:

    
fig_px1, ax_px1 = plt.subplots(nrows=2, ncols=4, figsize=(22, 10))

ax_px1[0, 0].imshow(dataset_vis1[0][1])
ax_px1[0, 0].set_xlabel('ISODATA')
ax_px1[0, 1].imshow(dataset_vis1[1][1])
ax_px1[0, 1].set_xlabel('Minimum')
ax_px1[0, 2].imshow(dataset_vis1[2][1])
ax_px1[0, 2].set_xlabel('Yen')
ax_px1[0, 3].imshow(dataset_vis1[3][1])
ax_px1[0, 3].set_xlabel('Mean')
ax_px1[1, 0].imshow(dataset_vis1[4][1])
ax_px1[1, 0].set_xlabel('Triangle')
ax_px1[1, 1].imshow(dataset_vis1[5][1])
ax_px1[1, 1].set_xlabel('Li')
ax_px1[1, 2].imshow(dataset_vis1[6][1])
ax_px1[1, 2].set_xlabel('Otsu')
ax_px1[1, 3].imshow(dataset_vis1[7][1])
ax_px1[1, 3].set_xlabel('MLSOS')

plt.show()



In [11]:

    
# normalizing TP, TN, FP and FN for each threshold result.
all_px1 = np.prod(img_gt1.shape)

dataset_normal1 = []

for data in dataset_vis1:
    dataset_normal1.append(np.asanyarray(data[2]) / all_px1)

pt_colors = ['g', 'b', 'r', '#ffff33']
pos = [1, 1.25, 1.5, 1.75]
mean_px1 = np.zeros(4)
dataset_normal1 = np.asanyarray(dataset_normal1)
num_boxes = len(dataset_normal1[0])
medians = list(range(num_boxes))

for idx, _ in enumerate(mean_px1):
    mean_px1[idx] = dataset_normal1[:, idx].mean()

labels = ['True\nPositive', 'True\nNegative', 'False\nPositive', 'False\nNegative']
width = (0.05, 0.05, 0.05, 0.05)

flier_props = dict(marker=(8, 2, 0.1), markerfacecolor='m', markersize=8,
                   markeredgecolor='m', linestyle='none')

fig, ax = plt.subplots(figsize=(16, 14))
box_plot = ax.boxplot(dataset_normal1, flierprops=flier_props, widths=width, positions=pos)

for i in range(num_boxes):
    box = box_plot['boxes'][i]
    boxX, boxY = [[] for _ in range(2)]
    for j in range(5):
        boxX.append(box.get_xdata()[j])
        boxY.append(box.get_ydata()[j])
    box_coords = list(zip(boxX, boxY))
    box_polygon = mpatches.Polygon(box_coords,
                                   facecolor='0.9')
    ax.add_patch(box_polygon)

    med = box_plot['medians'][i]
    medianX, medianY = [[] for _ in range(2)]
    for j in range(2):
        medianX.append(med.get_xdata()[j])
        medianY.append(med.get_ydata()[j])
        plt.plot(medianX, medianY, 'k')
        medians[i] = medianY[0]

ax.set_xticklabels(labels)
plt.axis([0.85, 1.9, -0.025, 1])

for idx, px in enumerate(pos):
    ax.plot(px, dataset_normal1[-1][idx],
            'o', markersize=8, markeredgecolor='k',
            color=pt_colors[idx])
ax.plot(pos, mean_px1, 'P', markersize=8, color='k', markeredgecolor='k')
plt.show()

Segmentation and comparison for test image 2.

Obtaining each threshold.



In [12]:

    
dataset_bin2 = [img_test2 < threshold_isodata(img_test2),
                img_test2 < threshold_minimum(img_test2),
                img_test2 < threshold_yen(img_test2),
                img_test2 < threshold_mean(img_test2),
                img_test2 < threshold_triangle(img_test2),
                img_test2 < threshold_li(img_test2),
                img_test2 < threshold_otsu(img_test2),
                img_mlsosbin2]

Obtaining the SSIM for each threshold.



In [13]:

    
label_info2 = [' * ISODATA X GT: ', ' * Minimum X GT: ', ' * Yen X GT: ',
               ' * Mean X GT: ', ' * Triangle X GT: ', ' * Li X GT: ',
               ' * Otsu X GT: ', ' * MLSOS X GT: ']

print('SSIM - test2.jpg.')

for idx, img_bin2 in enumerate(dataset_bin2):
    print(label_info2[idx], compare_ssim(img_bin2, img_gt2))









    



SSIM - test2.jpg.
 * ISODATA X GT:  0.256758040247
 * Minimum X GT:  0.878675667981
 * Yen X GT:  0.961813334452
 * Mean X GT:  0.327576126922
 * Triangle X GT:  0.463485049384
 * Li X GT:  0.246249068033
 * Otsu X GT:  0.256758040247
 * MLSOS X GT:  0.853944739317

Visual comparison between the thresholds and the GT for image 2.



In [14]:

    
dataset_vis2 = []

for idx, img_bin2 in enumerate(dataset_bin2):
    dataset_vis2.append(visual_comparison(img_bin2, img_gt2))



In [15]:

    
fig_comp2, ax_comp2 = plt.subplots(nrows=2, ncols=4, figsize=(22, 10))

ax_comp2[0, 0].imshow(dataset_vis2[0][0])
ax_comp2[0, 0].set_xlabel('ISODATA')
ax_comp2[0, 1].imshow(dataset_vis2[1][0])
ax_comp2[0, 1].set_xlabel('Minimum')
ax_comp2[0, 2].imshow(dataset_vis2[2][0])
ax_comp2[0, 2].set_xlabel('Yen')
ax_comp2[0, 3].imshow(dataset_vis2[3][0])
ax_comp2[0, 3].set_xlabel('Mean')
ax_comp2[1, 0].imshow(dataset_vis2[4][0])
ax_comp2[1, 0].set_xlabel('Triangle')
ax_comp2[1, 1].imshow(dataset_vis2[5][0])
ax_comp2[1, 1].set_xlabel('Li')
ax_comp2[1, 2].imshow(dataset_vis2[6][0])
ax_comp2[1, 2].set_xlabel('Otsu')
ax_comp2[1, 3].imshow(dataset_vis2[7][0])
ax_comp2[1, 3].set_xlabel('MLSOS')

plt.show()



In [16]:

    
fig_px2, ax_px2 = plt.subplots(nrows=2, ncols=4, figsize=(22, 10))

ax_px2[0, 0].imshow(dataset_vis2[0][1])
ax_px2[0, 0].set_xlabel('ISODATA')
ax_px2[0, 1].imshow(dataset_vis2[1][1])
ax_px2[0, 1].set_xlabel('Minimum')
ax_px2[0, 2].imshow(dataset_vis2[2][1])
ax_px2[0, 2].set_xlabel('Yen')
ax_px2[0, 3].imshow(dataset_vis2[3][1])
ax_px2[0, 3].set_xlabel('Mean')
ax_px2[1, 0].imshow(dataset_vis2[4][1])
ax_px2[1, 0].set_xlabel('Triangle')
ax_px2[1, 1].imshow(dataset_vis2[5][1])
ax_px2[1, 1].set_xlabel('Li')
ax_px2[1, 2].imshow(dataset_vis2[6][1])
ax_px2[1, 2].set_xlabel('Otsu')
ax_px2[1, 3].imshow(dataset_vis2[7][1])
ax_px2[1, 3].set_xlabel('MLSOS')

plt.show()



In [17]:

    
# normalizing TP, TN, FP and FN for each threshold result.
all_px2 = np.prod(img_gt2.shape)

dataset_normal2 = []

for data in dataset_vis2:
    dataset_normal2.append(np.asanyarray(data[2]) / all_px2)

# obtaining the mean for each column.
mean_px2 = np.zeros(4)

pt_colors = ['g', 'b', 'r', '#ffff33']
pos = [1, 1.25, 1.5, 1.75]
mean_px2 = np.zeros(4)
dataset_normal2 = np.asanyarray(dataset_normal2)
num_boxes = len(dataset_normal2[0])
medians = list(range(num_boxes))

for idx, _ in enumerate(mean_px2):
    mean_px2[idx] = dataset_normal2[:, idx].mean()

labels = ['True\nPositive', 'True\nNegative', 'False\nPositive', 'False\nNegative']
width = (0.05, 0.05, 0.05, 0.05)

flier_props = dict(marker=(8, 2, 0.1), markerfacecolor='m', markersize=8,
                   markeredgecolor='m', linestyle='none')

fig, ax = plt.subplots(figsize=(16, 14))
box_plot = ax.boxplot(dataset_normal2, flierprops=flier_props, widths=width, positions=pos)

for i in range(num_boxes):
    box = box_plot['boxes'][i]
    boxX, boxY = [[] for _ in range(2)]
    for j in range(5):
        boxX.append(box.get_xdata()[j])
        boxY.append(box.get_ydata()[j])
    box_coords = list(zip(boxX, boxY))
    box_polygon = mpatches.Polygon(box_coords,
                                   facecolor='0.9')
    ax.add_patch(box_polygon)

    med = box_plot['medians'][i]
    medianX, medianY = [[] for _ in range(2)]
    for j in range(2):
        medianX.append(med.get_xdata()[j])
        medianY.append(med.get_ydata()[j])
        plt.plot(medianX, medianY, 'k')
        medians[i] = medianY[0]

ax.set_xticklabels(labels)
plt.axis([0.85, 1.9, -0.025, 1])

for idx, px in enumerate(pos):
    ax.plot(px, dataset_normal2[-1][idx],
            'o', markersize=8, markeredgecolor='k',
            color=pt_colors[idx])
ax.plot(pos, mean_px2, 'P', markersize=8, color='k', markeredgecolor='k')
plt.show()



In [ ]:

Jansen-MIDAS: a multi-level photomicrograph segmentation software based on isotropic undecimated wavelets

Alexandre Fioravante de Siqueira1, 2, Flávio Camargo Cabrera2, Wagner Massayuki Nakasuga1, 2, Aylton Pagamisse3, Aldo Eloizo Job2

Importing necessary packages.

Setting plot variables.

Reading test images test1.jpg and test2.jpg.

Reading the ground truth of the test images, test1_GT.jpg and test2_GT.jpg.

Reading the MLSOS result of the test images, test1_MLSOSL03L6.tif and test2_MLSOSL03L6.tif.

Segmentation and comparison for test image 1.

Obtaining each threshold.

Obtaining SSIM for each threshold.

Visual comparison between the thresholds and the GT for the image 1.

Segmentation and comparison for test image 2.

Obtaining each threshold.

Obtaining the SSIM for each threshold.

Visual comparison between the thresholds and the GT for image 2.

Alexandre Fioravante de Siqueira^{1, 2}, Flávio Camargo Cabrera², Wagner Massayuki Nakasuga^{1, 2}, Aylton Pagamisse³, Aldo Eloizo Job²

Reading test images `test1.jpg` and `test2.jpg`.

Reading the ground truth of the test images, `test1_GT.jpg` and `test2_GT.jpg`.

Reading the MLSOS result of the test images, `test1_MLSOSL03L6.tif` and `test2_MLSOSL03L6.tif`.