CODEWITHSUNDEEP
pythonopencvcomputer-visionellipse
dcoppin
dcoppin

Reputation: 31

cv2.fitEllipse creates an ellipse totally different from what it should be

I want to fit ellipses around different clusters. Until now, I used cv2.fitEllipse and it worked fine. But I discovered a problem with one ellipse (out of more than 100). Here is a sample of code that reproduces the unexpected behavior.

import numpy as np
import cv2
import itertools
import matplotlib.pyplot as mpl

test=np.zeros((40,40))
test[27,12]=1
test[28,11:13]=1
test[29,11:14]=1
test[30,10:14]=1
test[31,11:14]=1
test[32,12:14]=1
test[33,13]=1

def getContours(clusters):
        map_contour=np.ones(clusters.shape).astype(np.uint8)
        int_cnts=[]
        for label in np.unique(clusters):
                # if the label is zero, we are examining the 'background'
                if label == 0:
                        continue
                mask = np.zeros(test.shape, dtype="uint8")
                mask[clusters == label] = 255
                cnts, hierarchy = \
                        cv2.findContours(mask.copy(),cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)
                int_cnts.append(cnts)
                c = max(cnts, key=cv2.contourArea)
                cv2.drawContours(map_contour, cnts, 0, (0, 255, 0), 1)
        list_contours=list(itertools.chain.from_iterable(int_cnts))
        return list_contours, map_contour

list_test,map_test=getContours(test)

info=np.zeros((len(list_test),5))
map_ellipse=np.ones(test.shape).astype(np.uint8)
for index,cnt in enumerate(list_test):
        ellipse=cv2.fitEllipse(cnt)
        info[index,0:2],info[index,2:4],info[index,4] = ellipse
        cv2.ellipse(map_ellipse,ellipse,(255,0,0))

print cnt
print ellipse

mpl.contourf(map_ellipse)
mpl.contour(test)
mpl.show()

cnt gives:

array([[[12,  27]],

       [[11,  28]],

       [[11,  29]],

       [[10, 30]],

       [[13, 33]],

       [[13,  29]],

       [[12,  28]]], dtype=int32)

ellipse is:

((23.71430015563965, -20.397018432617188),
 (8.957982063293457, 98.63383483886719),
 0.0)

As you can see if you copy/paste and run the code above, the top of the ellipse (cut in half) appears to be totally disconnected from the zone that it is supposed to represent (top left). I have no idea what is wrong is the code, especially since it works for other forms in the same picture.

What is causing this issue?

Edit: It seems that the ellipse is much better placed and sized (although not good enough) when I set test[30,9:14]=1 instead of test[30,10:14]=1. But when I use test[30,9:14]=1 instead, the ellipse is now bigger again and located in the top left corner. Such a sensitivity to one pixel that doesn't really change the shape of the form is outstanding.

Upvotes: 0

Views: 258

Answers (1)

dcoppin
dcoppin

Reputation: 31

I finally found what was going on. Because of the approximation cv2.CHAIN_APPROX_SIMPLE, the number of points used to calculate the ellipse was too small.

The problem is solved if I use cv2.CHAIN_APPROX_NONE, which takes into account all the points of contour.

Upvotes: 1

Related Questions