Notebook Demonstration of OpenCV's cv2 API for manipulating images, creating contours, simplifying those contours and creating various bounding structures. I started with Abid Rahman's Contours 1 and 2 blog posts. Those posts do not directly give you all that you need to run them. By virtue of being an IPython Notebook, these are the actual command that worked for me. You will need wget, numpy, matplotlib, OpenCV and shapely. -kurt schwehr 2013-Jan-31

In [ ]:

# BEGIN http://www.opencvpython.blogspot.com/2012/06/hi-this-article-is-tutorial-which-try.html

In [1]:

!wget http://3.bp.blogspot.com/-a5blM3JLkIU/T9OXg1YhN0I/AAAAAAAAASQ/MbdfSG2oaYg/s200/test.jpg

--2013-01-31 14:21:03--  http://3.bp.blogspot.com/-a5blM3JLkIU/T9OXg1YhN0I/AAAAAAAAASQ/MbdfSG2oaYg/s200/test.jpg
Resolving 3.bp.blogspot.com (3.bp.blogspot.com)... 2001:4860:4001:802::1008, 74.125.224.70, 74.125.224.72, ...
Connecting to 3.bp.blogspot.com (3.bp.blogspot.com)|2001:4860:4001:802::1008|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 2605 (2.5K) [image/jpeg]
Saving to: `test.jpg'

100%[======================================>] 2,605       --.-K/s   in 0s      

2013-01-31 14:21:03 (292 MB/s) - `test.jpg' saved [2605/2605]

In [92]:

import numpy as np
import cv2
import cv
import shapely.geometry

In [3]:

im = cv2.imread('test.jpg')

In [11]:

whos

Variable    Type       Data/Info
--------------------------------
contours    list       n=1
cv          module     <module 'cv' from '/usr/l<...>odules/python2.7/cv.pyc'>
cv2         module     <module 'cv2' from '/usr/<...>odules/python2.7/cv2.so'>
hierarchy   ndarray    1x1x4: 4 elems, type `int32`, 16 bytes
im          ndarray    200x200x3: 120000 elems, type `uint8`, 120000 bytes (117 kb)
imgray      ndarray    200x200: 40000 elems, type `uint8`, 40000 bytes
ret         float      127.0
thresh      ndarray    200x200: 40000 elems, type `uint8`, 40000 bytes

In [7]:

imgray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)

In [14]:

ret, thresh = cv2.threshold(imgray, 127, 255, 0)
imshow(thresh)

Out[14]:

<matplotlib.image.AxesImage at 0x42cfe50>

In [15]:

contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
imshow(thresh)

Out[15]:

<matplotlib.image.AxesImage at 0x4467dd0>

In [12]:

len(contours), hierarchy

Out[12]:

(1, array([[[-1, -1, -1, -1]]], dtype=int32))

In [17]:

len(contours[0])

Out[17]:

In [18]:

cv2.drawContours(im, contours, -1, (0,255,0), 3)
imshow(im)

Out[18]:

<matplotlib.image.AxesImage at 0x4855350>

In [19]:

cv2.drawContours(im, contours, -1, (0,255,0), -1)
imshow(im)

Out[19]:

<matplotlib.image.AxesImage at 0x4a61850>

In [22]:

cv2.drawContours(im, contours, 0, (0,255,0), 1)
imshow(im)

Out[22]:

<matplotlib.image.AxesImage at 0x4a96650>

In [23]:

!wget http://3.bp.blogspot.com/-1UtLXb7c73U/T9QZT3tpVjI/AAAAAAAAATE/Nyo7SFg8T1o/s1600/balls.png

--2013-01-31 15:06:48--  http://3.bp.blogspot.com/-1UtLXb7c73U/T9QZT3tpVjI/AAAAAAAAATE/Nyo7SFg8T1o/s1600/balls.png
Resolving 3.bp.blogspot.com (3.bp.blogspot.com)... 2001:4860:4001:800::1005, 74.125.224.97, 74.125.224.101, ...
Connecting to 3.bp.blogspot.com (3.bp.blogspot.com)|2001:4860:4001:800::1005|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 3499 (3.4K) [image/png]
Saving to: `balls.png'

100%[======================================>] 3,499       --.-K/s   in 0s      

2013-01-31 15:06:48 (337 MB/s) - `balls.png' saved [3499/3499]

In [24]:

im = cv2.imread('balls.png')
imgray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)

In [28]:

ret, thresh = cv2.threshold(imgray, 100, 255, 0)
imshow(thresh)

Out[28]:

<matplotlib.image.AxesImage at 0x57eb390>

In [30]:

contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

In [31]:

for h, cnt in enumerate(contours):
    mask = np.zeros(imgray.shape, np.uint8)
    cv2.drawContours(mask, [cnt], 0, 255, -1)
    mean = cv2.mean(im, mask=mask)
    print h, mean

0 (0.0, 242.0, 255.0, 0.0)
1 (232.0, 162.0, 0.0, 0.0)
2 (76.0, 177.0, 34.0, 0.0)
3 (39.0, 127.0, 255.0, 0.0)

In [32]:

# END http://www.opencvpython.blogspot.com/2012/06/hi-this-article-is-tutorial-which-try.html

BEGIN Contours - 2¶

In [37]:

moments = cv2.moments(contours[0])
area = moments['m00']
moments

Out[37]:

{'m00': 10033.5,
 'm01': 1850134.5,
 'm02': 352044973.5833333,
 'm03': 68983799276.15001,
 'm10': 2010061.8333333333,
 'm11': 368666693.125,
 'm12': 69763579350.98334,
 'm20': 409360323.5833333,
 'm21': 74691021944.88333,
 'm30': 84692116672.95001,
 'mu02': 10888082.359906793,
 'mu03': 0.005234025965704581,
 'mu11': -1980114.5675549507,
 'mu12': -33122544.260385513,
 'mu20': 6674463.831166983,
 'mu21': 101313.30416250229,
 'mu30': 8633090.369003296,
 'nu02': 0.10815497152071127,
 'nu11': -0.019669141689288665,
 'nu12': -0.0032846761082870463,
 'nu20': 0.06629968636479547,
 'nu21': 1.0046975468372928e-05,
 'nu30': 0.0008561209988226333}

In [38]:

cv2.contourArea(contours[0])

Out[38]:

10033.5

In [39]:

perimeter = cv2.arcLength(contours[0], True) # True says that curve is closed
perimeter

Out[39]:

473.34523379802704

In [42]:

cnt = contours[0]
len(cnt)

Out[42]:

In [57]:

x = cnt[:,0, 0]; y = cnt[:,0,1]
plot(x,y)

Out[57]:

[<matplotlib.lines.Line2D at 0x5ac8f50>]

In [61]:

simpler = cv2.approxPolyDP(cnt, 2, True)
plot(simpler[:,0,0], simpler[:,0,1])

Out[61]:

[<matplotlib.lines.Line2D at 0x60c09d0>]

In [63]:

hull = cv2.convexHull(cnt)
plot(hull[:,0,0], hull[:,0,1])

Out[63]:

[<matplotlib.lines.Line2D at 0x6381fd0>]

In [69]:

simpler_hull = cv2.approxPolyDP(hull, 2.5, True)
plot(simpler_hull[:,0,0], simpler_hull[:,0,1])
plot(x,y)

Out[69]:

[<matplotlib.lines.Line2D at 0x6957e10>]

In [73]:

r_x, r_y, r_w, r_h = cv2.boundingRect(cnt)
r_x, r_y, r_w, r_h  # (150, 121, 103, 146)
plot((r_x, r_x, r_x+r_w, r_x+r_w, r_x), (r_y, r_y+r_h, r_y+r_h, r_y, r_y))
plot(x,y)

Out[73]:

[<matplotlib.lines.Line2D at 0x6eab610>]

In [88]:

rect = cv2.minAreaRect(cnt)
rect # ((202.134521484375, 192.14178466796875), (102.39618682861328, 140.3079376220703), -5.128190994262695)
box = cv2.cv.BoxPoints(rect)
box # ((157.41201782226562, 266.59124755859375), (144.87069702148438, 126.84494018554688), (246.85702514648438, 117.69232177734375), (259.3983459472656, 257.4386291503906))
# plot( [p[0] for p in box] + [box[0][0]], [p[1] for p in box] + [box[0][1]] )
box_list = list(box)
box_list.append(box[0])
ba = np.array(box_list) # Box array
plot(ba[:,0], ba[:,1])
plot(x,y)

Out[88]:

[<matplotlib.lines.Line2D at 0x7746a10>]

In [97]:

(c_x, c_y), radius = cv2.minEnclosingCircle(cnt)
c_x, c_y, radius # (197.0, 194.5, 82.92139434814453)
center = shapely.geometry.Point(c_x, c_y)
circle = np.array(center.buffer(radius).boundary.coords)
len(circle) # 66 points
plot(circle[:,0], circle[:,1])
plot(x,y)

Out[97]:

[<matplotlib.lines.Line2D at 0x7ba81d0>]

In [100]:

ellipse = cv2.fitEllipse(cnt)
ellipse # ((199.31251525878906, 185.9192352294922), (93.7149658203125, 138.58531188964844), 202.948486328125)
# TODO: what is a convenient way to get the coords for an ellipse?

Out[100]:

((199.31251525878906, 185.9192352294922),
 (93.7149658203125, 138.58531188964844),
 202.948486328125)

In [ ]: