Why does my code not segment the characters of the input image?
I am trying to do character segmentation from an image, but the output is not getting segmented.
That's my new_input image:
That's my code for segmenting characters:
word = cv2.imread('../input/word-image/word.PNG',0)
fig, ax1 = plt.subplots(1)
ax1.imshow(word, cmap="gray")
# the next two lines is based on the assumptions that the width of
# a license plate should be between 5% and 15% of the license plate,
# and height should be between 35% and 60%
# this will eliminate some
character_dimensions = (0.35*word.shape[0], 0.60*word.shape[0], 0.05*word.shape[1], 0.15*word.shape[1])
min_height, max_height, min_width, max_width = character_dimensions
characters = []
counter=0
column_list = []
for regions in regionprops(word):
y0, x0, y1, x1 = regions.bbox
region_height = y1 - y0
region_width = x1 - x0
if region_height > min_height and region_height < max_height and region_width > min_width and region_width < max_width:
roi = word[y0:y1, x0:x1]
# draw a red bordered rectangle over the character.
rect_border = patches.Rectangle((x0, y0), x1 - x0, y1 - y0, edgecolor="red",
linewidth=2, fill=False)
ax1.add_patch(rect_border)
# resize the characters to 20X20 and then append each character into the characters list
resized_char = resize(roi, (20, 20))
characters.append(resized_char)
# this is just to keep track of the arrangement of the characters
column_list.append(x0)
# print(characters)
plt.show()
That's the current output:
What could be the problem here?
From the documentation on skimage.measure.regionprops:
Measure properties of labeled image regions.
You don't feed a proper labeled image, but your input is interpreted as one, since you have several gray values due to the aliasing. Debug regions inside your loop, and you'll see, that a lot of regions are detected. They're all dismissed due to your assumptions on width and height.
So, the first step would be to generate a proper labeled image, e.g. using cv2.connectedComponents. Therefore, you'd need to (inverse) binarize your input image beforehand. Having the labels image, you can directly proceed with your loop. Nevertheless, I would dismiss all assumptions on width and height here.
That'd be the modified code:
import cv2
import matplotlib.pyplot as plt
from matplotlib import patches
from skimage.measure import regionprops
from skimage.transform import resize
# Read image as grayscale
word = cv2.imread('pFLpN.png', cv2.IMREAD_GRAYSCALE)
# Inverse binarize image, and find connected components
thr = cv2.threshold(word, 254, 255, cv2.THRESH_BINARY_INV)[1]
labels = cv2.connectedComponents(thr)[1]
# Maybe leave out any assumptions on the width and height...
# Prepare outputs
plt.figure(figsize=(18, 9))
plt.subplot(2, 2, 1), plt.imshow(word, cmap='gray'), plt.title('Original image')
plt.subplot(2, 2, 2), plt.imshow(thr, cmap='gray'), plt.title('Binarized image')
plt.subplot(2, 2, 3), plt.imshow(labels), plt.title('Connected components')
ax = plt.subplot(2, 2, 4), plt.imshow(word, cmap='gray')
# Iterate found connected components as before
# (Without checking width and height...)
characters = []
counter = 0
column_list = []
for regions in regionprops(labels):
y0, x0, y1, x1 = regions.bbox
region_height = y1 - y0
region_width = x1 - x0
roi = word[y0:y1, x0:x1]
rect_border = patches.Rectangle((x0, y0), x1 - x0, y1 - y0, edgecolor='red',
linewidth=2, fill=False)
ax[0].add_patch(rect_border)
resized_char = resize(roi, (20, 20))
characters.append(resized_char)
column_list.append(x0)
plt.title('Segmented characters')
plt.tight_layout(), plt.show()
And, that'd be the output:
----------------------------------------
System information
----------------------------------------
Platform: Windows-10-10.0.19041-SP0
Python: 3.9.1
PyCharm: 2021.1.2
Matplotlib: 3.4.2
OpenCV: 4.5.2
scikit-image: 0.18.1
----------------------------------------