CODEWITHSUNDEEP
pythonopencv
Kimwaga Makono
Kimwaga Makono

Reputation: 121

How to do remove jagged edges of an object in an image with python and opencv

I have tried to search and found two solutions that could solve the problem I have but are not working by my side. I want to remove the jagged edges or smooth the image attached. I have tried to implement the following codes 

import cv2
import numpy as np
import os
os.chdir("F:/Examples")
image = cv2.imread("image1.jpeg")
blur = cv2.GaussianBlur(image, (21, 21), 0)
mask = np.zeros(image.shape[:2], dtype=np.uint8)

output = np.where(mask==np.array([255, 255, 255]), blur, image)
cv2.imwrite("", output)

The above codes returns unchanged image as if the applied mask didn't work. Here is the input image

I also tried to implement some solutions from different links including (How to blur the image according to segmentation map) without success. Any help will be acknowledged While the edges or the external contour should be clear as in this image

Upvotes: 2

Views: 3107

Answers (1)

Rotem
Rotem

Reputation: 32084

I almost managed to solve it.

The main problem is that the contour is partially curved, and partially with straight lines.

Explanations are in the comments:

import numpy as np
import cv2
from scipy.interpolate import splprep, splev

im = cv2.imread('image1.jpeg')

bk = im.copy()

# Fill background with black color
cv2.floodFill(im, None, seedPoint=(1,1), newVal=(0, 0, 0), loDiff=(5, 5, 5), upDiff=(5, 5, 5))
gray = cv2.cvtColor(im, cv2.COLOR_RGB2GRAY)

ret, thresh_gray = cv2.threshold(gray, 5, 255, cv2.THRESH_BINARY)

# Use "open" morphological operation for removing small contours (noise)
thresh_gray = cv2.morphologyEx(thresh_gray, cv2.MORPH_OPEN, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5,5)));

# Background
bk[(thresh_gray > 0)] = 0
bk = cv2.morphologyEx(bk, cv2.MORPH_DILATE, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20,20)));
#cv2.imshow('bk', bk)

# Foreground
fg = im.copy()
tmm_fg = cv2.morphologyEx(fg, cv2.MORPH_DILATE, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20,20)));
fg_gray = cv2.cvtColor(fg, cv2.COLOR_RGB2GRAY)
fg[(fg_gray==0)] = tmm_fg[(fg_gray==0)]


#thresh_gray = cv2.morphologyEx(thresh_gray, cv2.MORPH_CLOSE, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (50,50)));

# Find contours (there is only one contour)
# _, contours, _ = cv2.findContours(thresh_gray, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)  # 3 outputs works only in OpenCV 3. [-2:] is used for compatibility.
contours, _ = cv2.findContours(thresh_gray, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)[-2:]  # https://stackoverflow.com/questions/48291581/how-to-use-cv2-findcontours-in-different-opencv-versions
c = contours[0]

# Smooth contour
# https://agniva.me/scipy/2016/10/25/contour-smoothing.html
x,y = c.T
x = x.tolist()[0]
y = y.tolist()[0]
tck, u = splprep([x,y], u=None, s=1.0, per=1)
u_new = np.linspace(u.min(), u.max(), 150)
x_new, y_new = splev(u_new, tck, der=0)
res_array = [[[int(i[0]), int(i[1])]] for i in zip(x_new,y_new)]
smoothened = np.asarray(res_array, dtype=np.int32)

# Build a mask
mask = np.zeros_like(thresh_gray)
cv2.drawContours(mask, [smoothened], -1, 255, -1)

# For testig
test_im = cv2.cvtColor(thresh_gray, cv2.COLOR_GRAY2RGB)
cv2.drawContours(test_im, [smoothened], 0, (0, 255, 0), 1)

res = bk
res[(mask > 0)] = fg[(mask > 0)]

cv2.imshow('test_im', test_im)
cv2.imshow('res', res)
cv2.waitKey(0)
cv2.destroyAllWindows()

Upvotes: 2

Related Questions