Make Edges of the Image smooth

Question

I am currently working a simple project
It is removing the Background of any image and converting it into a Sticker but it is not Giving me Smoother

import cv2
import numpy as np
from PIL import Image, ImageFilter
from google.colab.patches import cv2_imshow
from matplotlib import pyplot as pl
#img = cv2.imread("/content/police-car-icon-cartoon-style-vector-16884775.jpg")
remove_background("/content/WhatsApp Image 2020-08-17 at 1.08.33 AM (2).jpeg")




def remove_background(img1):

#== Parameters =======================================================================

BLUR = 5
CANNY_THRESH_1 = 10
CANNY_THRESH_2 = 100
MASK_DILATE_ITER = 10
MASK_ERODE_ITER = (1,1)
MASK_COLOR = (220,220,220) # In BGR format

#== Processing =======================================================================

#-- Read image -----------------------------------------------------------------------
img = cv2.imread(img1)
#img = cv2.resize(img, (600,600))
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

#-- Edge detection -------------------------------------------------------------------
edges = cv2.Canny(gray, CANNY_THRESH_1, CANNY_THRESH_2)
edges = cv2.dilate(edges, None)
##edges = cv2.erode(edges, None)

#-- Find contours in edges, sort by area ---------------------------------------------
contour_info = []
contours, _ = cv2.findContours(edges, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)

for c in contours:
    contour_info.append((
        c,
        cv2.isContourConvex(c),
        cv2.contourArea(c),
    ))
contour_info = sorted(contour_info, key=lambda c: c[2], reverse=True)


#-- Create empty mask, draw filled polygon on it corresponding to largest contour ----
# Mask is black, polygon is white
mask = np.zeros(edges.shape)
for c in contour_info:
    cv2.fillConvexPoly(mask, c[0], (255))
# cv2.fillConvexPoly(mask, max_contour[0], (255))

#-- Smooth mask, then blur it --------------------------------------------------------
mask = cv2.dilate(mask, None, iterations=MASK_DILATE_ITER)
mask_stack = np.dstack([mask]*3)    # Create 3-channel alpha mask

mask_u8 = np.array(mask,np.uint8)

back = np.zeros(mask.shape,np.uint8)
back[mask_u8 == 0] = 255

border = cv2.Canny(mask_u8, CANNY_THRESH_1, CANNY_THRESH_2)
border = cv2.dilate(border, None, iterations=3)


masked = mask_stack * img  # Blend
masked = (masked * 255).astype('uint8')

#     background Colors (blue,green,red)
masked[:,:,0][back == 255] = 190
masked[:,:,1][back == 255] = 190
masked[:,:,2][back == 255] = 190





cv2.imwrite('img.png', masked)

cv2_imshow(  masked)

cv2.waitKey(0)
cv2.destroyAllWindows()

This is the Output Image

But I want this image to be little smoother like this

Answer 1

Here is how to replace the background with some colored image rather than transparency in Python/OpenCV.

• Read the input
• Convert to gray
• Threshold
• Blur then stretch gray to black to antialias
• Get the external contours and the largest contour
• Draw the largest contour as white on black background
• Dilate to add your black border (if desired)
• Create a colored (red) background image
• Apply the mask to the input
• Apply the inverted mask to the background
• Add the two results together
• Save the result

Input:

import cv2
import numpy as np
import skimage.exposure

# load image
img = cv2.imread('bunny.jpg')

# convert to gray
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# threshold
thresh = cv2.threshold(gray, 32, 255, cv2.THRESH_BINARY)[1]

# blur threshold image
blur = cv2.GaussianBlur(thresh, (0,0), sigmaX=3, sigmaY=3, borderType = cv2.BORDER_DEFAULT)

# stretch so that 255 -> 255 and 127.5 -> 0
stretch = skimage.exposure.rescale_intensity(blur, in_range=(127.5,255), out_range=(0,255)).astype(np.uint8)

# threshold again
thresh2 = cv2.threshold(stretch, 0, 255, cv2.THRESH_BINARY)[1]

# get external contour
contours = cv2.findContours(thresh2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]
big_contour = max(contours, key=cv2.contourArea)

# draw white filled contour on black background
contour = np.zeros_like(thresh, dtype=np.uint8)
cv2.drawContours(contour, [big_contour], 0, 255, -1)

# dilate mask for dark border
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20,20))
mask = cv2.morphologyEx(contour, cv2.MORPH_DILATE, kernel)

# create red colored background image
bckgrnd = np.full_like(img, (0,0,255), dtype=np.uint8)

# apply mask to img
img_masked = cv2.bitwise_and(img, img, mask=mask)

# apply inverse mask to colored background image
bckgrnd_masked = cv2.bitwise_and(bckgrnd, bckgrnd, mask=255-mask)

# combine the two
result = cv2.add(img_masked, bckgrnd_masked)

# save output
cv2.imwrite('bunny_thresh2.png', thresh)
cv2.imwrite('bunny_mask2.png', mask)
cv2.imwrite('bunny_masked2.png', img_masked)
cv2.imwrite('bunny_background_masked2.png', bckgrnd_masked)
cv2.imwrite('bunny_result2.png', result)

# Display various images to see the steps
cv2.imshow('gray',gray)
cv2.imshow('thresh', thresh)
cv2.imshow('blur', blur)
cv2.imshow('stretch', stretch)
cv2.imshow('thresh2', thresh2)
cv2.imshow('contour', contour)
cv2.imshow('mask', mask)
cv2.imshow('img_masked', img_masked)
cv2.imshow('bckgrnd_masked', bckgrnd_masked)
cv2.imshow('result', result)

cv2.waitKey(0)
cv2.destroyAllWindows()

Threshold image:

Mask image:

Mask applied to image:

Inverted mask applied to background:

Result:

Answer 2

Here is one way to do alpha channel antialiasing in Python/OpenCV

• Read the input
• Convert to grayscale
• threshold to create a mask
• blur
• Stretch contrast so mid gray goes to black
• Threshold again
• Get external contour
• Draw white filled contour on black background
• Dilate for dark border
• Blur again slightly
• Stretch contrast so mid gray goes to black as mask
• Put the mask into the alpha channel of the input
• Save results

Input:

import cv2
import numpy as np
import skimage.exposure

# load image
img = cv2.imread('bunny.jpg')

# convert to gray
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# threshold
thresh = cv2.threshold(gray, 32, 255, cv2.THRESH_BINARY)[1]

# blur threshold image
blur = cv2.GaussianBlur(thresh, (0,0), sigmaX=3, sigmaY=3, borderType = cv2.BORDER_DEFAULT)

# stretch so that 255 -> 255 and 127.5 -> 0
stretch = skimage.exposure.rescale_intensity(blur, in_range=(127.5,255), out_range=(0,255)).astype(np.uint8)

# threshold again
thresh2 = cv2.threshold(stretch, 0, 255, cv2.THRESH_BINARY)[1]

# get external contour
contours = cv2.findContours(thresh2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]
big_contour = max(contours, key=cv2.contourArea)

# draw white filled contour on black background as mas
contour = np.zeros_like(gray)
cv2.drawContours(contour, [big_contour], 0, 255, -1)

# dilate mask for dark border
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20,20))
dilate = cv2.morphologyEx(contour, cv2.MORPH_DILATE, kernel)

# blur dilate image
blur2 = cv2.GaussianBlur(dilate, (3,3), sigmaX=0, sigmaY=0, borderType = cv2.BORDER_DEFAULT)

# stretch so that 255 -> 255 and 127.5 -> 0
mask = skimage.exposure.rescale_intensity(blur2, in_range=(127.5,255), out_range=(0,255))

# put mask into alpha channel of input
result = cv2.cvtColor(img, cv2.COLOR_BGR2BGRA)
result[:,:,3] = mask

# save output
cv2.imwrite('bunnyman_thresh.png', thresh)
cv2.imwrite('bunny_mask.png', mask)
cv2.imwrite('bunny_antialiased.png', result)


# Display various images to see the steps
cv2.imshow('gray',gray)
cv2.imshow('thresh', thresh)
cv2.imshow('blur', blur)
cv2.imshow('stretch', stretch)
cv2.imshow('thresh2', thresh2)
cv2.imshow('contour', contour)
cv2.imshow('dilate', dilate)
cv2.imshow('mask', mask)
cv2.imshow('result', result)

cv2.waitKey(0)
cv2.destroyAllWindows()

Threshold image:

Mask image:

Result:

Answer 3

Would a dithering algorithm work? Here is a PIL extension for dithering: https://github.com/hbldh/hitherdither