PyDIP - use of MeasurementTool.Measure

[yg42]

Hello all !

I really like using PyDIP, though the documentation is sometimes difficult to find. For example, after segmentation (of a 2D image), I want to approximate each segment by an ellipsoid (mainly with minor and major axes). The, I did:

props = dip.MeasurementTool.Measure(lbl, features=['DimensionsEllipsoid'])

But I cannot figure out what the results returned ? Is this feature the right one to use ?

https://diplib.org/diplib-docs/features.html#binary_moments_DimensionsEllipsoid

Thank you for your help

[crisluengo]

It's the length of the small and large diameters of the best fit ellipsoid, in physical units if they were attached to the image (and the two dimensions have the same units), otherwise in pixels.

If you also want the angle of ellipsoid, use 'MajorAxes', which is the eigenvector matrix (4 values in 2D). So you get the two unit vectors corresponding to each of the two axes of the ellipsoid. Draw those vectors, each with a length corresponding to half of 'DimensionsEllipsoid', to see the two main radii of the ellipsoid.

'Center' is the centroid of the object, and thus also of the ellipsoid.

[yg42]

Thank you Cris !

To help other with this, here is the code I have now. It works on a 8 bits image.

import diplib as dip

# Load the image with opencv
I = dip.ImageRead('model.png')

image = dip.Image(I)>100
print(image)
# label the image
label_image = dip.Label(image)

# Use actual dimensions of pixels
length_pixel = dip.PhysicalQuantity(0.35, 'mm')

label_image.SetPixelSize(length_pixel)
props = dip.MeasurementTool.Measure(label_image, features=['DimensionsEllipsoid'])

print(props)

Which gives the following result:

2D Scalar image:
    data type:      BIN
    sizes:          {1000, 1000}
    strides:        {1, 1000}, tensor stride 1
    data pointer:   0x562ce9220160 (shared among 1 images)
    origin pointer: 0x562ce9220160

   |     DimensionsEllipsoid |                                   Mu | 
-- | ----------------------- | ------------------------------------ | 
   |      axis0 |      axis1 |     Mu_0_0 |     Mu_1_1 |     Mu_1_0 | 
   |       (mm) |       (mm) |      (mm²) |      (mm²) |      (mm²) | 
-- | ---------- | ---------- | ---------- | ---------- | ---------- | 
 1 |      5.971 |      5.849 |      2.138 |      2.229 |      0.000 | 
 2 |      8.468 |      3.452 |     0.7999 |      4.427 |    -0.4500 | 
 3 |      6.508 |      4.871 |      1.483 |      2.647 |      0.000 | 
 4 |      6.315 |      1.698 |     0.1801 |      2.492 |      0.000 | 
 5 |      4.419 |      1.872 |     0.2189 |      1.220 |  1.121e-14 | 
 6 |      6.017 |      1.700 |     0.1806 |      2.263 | -1.121e-14 | 
 7 |      6.251 |      6.251 |      2.442 |      2.442 |      0.000 | 
 8 |      19.73 |      8.705 |      7.545 |      21.52 |     -6.866 | 
 9 |      11.94 |      8.380 |      5.607 |      7.699 |     -2.008 | 
10 |      7.493 |      7.493 |      3.509 |      3.509 |      0.000 | 
11 |      6.555 |      6.555 |      2.685 |      2.685 |      0.000 | 
12 |      3.421 |      3.421 |     0.7315 |     0.7315 |      0.000 | 
13 |      7.493 |      7.493 |      3.509 |      3.509 |      0.000 | 
...
82 |      3.702 |      3.702 |     0.8567 |     0.8567 |      0.000 | 
83 |      4.854 |      3.869 |     0.9357 |      1.473 | -3.587e-13 | 
84 |      4.036 |      1.015 |    0.06437 |      1.018 |      0.000 | 

[crisluengo]

One thing to note is that the "Mu" feature is computed and added to the props output because the "DimensionsEllipsoid" feature is computed from it.