Extract ellipse shape food plate from image through Python

Question

Food Plate

The idea is to extract the plate which is in the shape of an ellipse.
I tried the HoughCircles method from OpenCV but it only works for perfect circles.
I also tried the hough_ellipse method from skimage but it is taking too long or I implemented it the wrong way.
Is it possible to detect ellipse shape using the OpenCV module?

What other solutions exist?

Food Plate:

Answer 1

The main key for extracting the plate is using cv2.adaptiveThreshold , but there are few more stages:

• Convert to Grayscale and apply adaptive threshold with relatively large Gaussian.
• Find connected components (clusters).
  Find largest cluster, and create new image with only the largest cluster.
• Use "open" morphological operation for removing some artifacts.
• Fill the plate with white pixels (using floodFill).
• Find contours, and get the contour with maximum area.
• Draw contour with maximum size to create a mask.
  Apply the mask on the original image.

Finding the ellipse by shape, is much less robust...

Here is the code:

import numpy as np
import cv2
import imutils

img = cv2.imread('food_plate.jpg')

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

# Apply adaptive threshold with gaussian size 51x51
thresh_gray = cv2.adaptiveThreshold(gray, 255, adaptiveMethod=cv2.ADAPTIVE_THRESH_GAUSSIAN_C, thresholdType=cv2.THRESH_BINARY, blockSize=51, C=0)

#cv2.imwrite('thresh_gray.png', thresh_gray)

# Find connected components (clusters)
nlabel,labels,stats,centroids = cv2.connectedComponentsWithStats(thresh_gray, connectivity=8)

# Find second largest cluster (the cluster is the background):
max_size = np.max(stats[1:, cv2.CC_STAT_AREA])
max_size_idx = np.where(stats[:, cv2.CC_STAT_AREA] == max_size)[0][0]

mask = np.zeros_like(thresh_gray)

# Draw the cluster on mask
mask[labels == max_size_idx] = 255

# Use "open" morphological operation for removing some artifacts
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5,5)))

#cv2.imwrite('mask.png', mask)

# Fill the plate with white pixels
cv2.floodFill(mask, None, tuple(centroids[max_size_idx].astype(int)), newVal=255, loDiff=1, upDiff=1)

#cv2.imwrite('mask.png', mask)

# Find contours, and get the contour with maximum area
cnts = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
cnts = imutils.grab_contours(cnts)

c = max(cnts, key=cv2.contourArea)

# Draw contours with maximum size on new mask
mask2 = np.zeros_like(mask)
cv2.drawContours(mask2, [c], -1, 255, -1)

#cv2.imwrite('mask2.png', mask2)

img[(mask2==0)] = 0

# Save result
cv2.imwrite('img.jpg', img)

Result: