[OpenCV]how to fix contours to rectangle? - opencv

Question

Note
I'm new to OpenCV(or computer vision), so it would be very helpful just to tell me the search query!
What I want to ask
I want to write a program that extract the business cards from pictures.
I was able to extract a rough outline, but reflected light becomes noise and I can't extract an accurate outline. Please tell me your idea.
image(raw data)
raw data
output
output data(rough outline)
code
import math
import itertools
from glob import glob
import cv2
import numpy as np
from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
%matplotlib inline
def read_images():
"read image data from data directory"
names = glob('data/*.jpg')
names.sort()
return map(lambda name: cv2.imread(name), names)
def blur(img):
"apply blur"
return cv2.GaussianBlur(img, (25, 25), 0)
def show_images(images, column, color_type=cv2.COLOR_BGR2RGB):
"plot images with matplotlib"
plt.figure(figsize=(10,10), dpi=150)
for n, img in zip(range(len(images)), images):
p = plt.subplot(math.ceil(len(images) / column), column, n + 1)
p.axis('off')
if color_type is None:
p.imshow(img)
else:
p.imshow(cv2.cvtColor(img, color_type))
plt.show()
def detect_background_color(img):
"detect background color"
# Assume that the perimeter is all background
height, width, *_ = img.shape
background_colors = np.concatenate([
img[5:height-5, 5], img[5, 5:width-5],
img[5:height-5, width-5], img[height-5, 5:width-5]
])
background_colors = background_colors.astype(np.float32)
# Assume that the background color is only one.
K = 2
iter_flg = cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER
_, labels, centers = cv2.kmeans(
background_colors, K, None, (iter_flg, 10, 1.0), 10,
cv2.KMEANS_RANDOM_CENTERS)
cnt1 = len(labels[labels==0])
cnt2 = len(labels[labels==1])
return centers[0] if cnt1 > cnt2 else centers[1]
def scale(img):
bg = detect_background_color(img)
return np.fix(np.sqrt(np.sum(np.square(img - bg), axis=2)) / 1.732).astype(np.uint8)
def binarize(img):
th, bit = cv2.threshold(img, 40, 255, cv2.THRESH_BINARY)
return bit
binarized = [binarize(scale(blur(img))) for img in read_images()]
show_images(binarized, 4, None)

Answer 1

Looks like you need to apply morphology try cv2.erode and then cv2 dilate operations.
The first will remove regions smaller than erode kernel size, the second will restore initial size of large blob. You need to apply the same size kernels for both operations.
morphology.
Check also this: medium article