Labeling and Measurement

Here we’re going to label a image and measure some features for each label.

Basics

Get the sample image.

import impy as ip
img = ip.sample_image("coins")

There are two methods for image labeling.

• label … labeling image with the input reference.

• label_threshold … labeling image with the binarized image as the reference. It can be considered as a shortcut of img.label(img.threshold()).

Both method returns a Label array, a subclass of MetaArray.

Simple labeling

img.label_threshold()

Label of
      name      : coins
     shape      : 303(y), 384(x)
     dtype      : uint8
     source     : None
     scale      : ScaleView(y=1.0, x=1.0)

The Label is tagged to the image itself.

img.labels

Label of
      name      : coins
     shape      : 303(y), 384(x)
     dtype      : uint8
     source     : None
     scale      : ScaleView(y=1.0, x=1.0)

Slicing of image covariantly slices the labels at the same time.

img[100:200, 53].labels

Label of
      name      : coins
     shape      : 100(y)
     dtype      : uint8
     source     : None
     scale      : ScaleView(y=1.0)

Image with label overlay can be visualized by imshow method or using napari viewer.

img.imshow(label=True)  # use matplotlib to show images
ip.gui.add(img)  # use napari to show images

Measurement

After labeling, regionprops method is useful for image measurement. This method runs skimage.measure.regionprops inside.

props = img.regionprops(properties=("mean_intensity", "area", "major_axis_length"))
props

DataDict[PropArray] with 3 components:
'mean_intensity' => PropArray of
      name      : coins-prop
     shape      : 98(N)
     dtype      : float32
     source     : None
 property name  : mean_intensity
,
'area' => PropArray of
      name      : coins-prop
     shape      : 98(N)
     dtype      : float32
     source     : None
 property name  : area
,
'major_axis_length' => PropArray of
      name      : coins-prop
     shape      : 98(N)
     dtype      : float32
     source     : None
 property name  : major_axis_length

The returned DataDict object is a dict-like object. Its value is assured to be the same type so that you can easily apply a same method to all the components (see ../tutorial.rst). Since “mean_intensity”, “area” and “major_axis_length” are chosen for measurement, props has keys “mean_intensity”, “area” and “major_axis_length”.

Here, props is a DataDict of PropArray. PropArray is a subclass of MetaArray that is specialized in storing properties.

All the properties can be summerized as follows.

# since PropArray has method `mean`, this line will apply `mean` to all the components.
props.mean()

DataDict[float32] with 3 components:
'mean_intensity' => 122.47181,
'area' => 465.52042,
'major_axis_length' => 15.488672

# PropArray has a visualization method `hist`.
props.hist()

Conditional Labeling

Simple labeling based on thresholding always yields in insufficient results.

The filt argument can filter labels based on properties of image and labels. Basic usage is following. Filter function must take at least two argument, image itself and newly created label region.

def filt(img, lbl):
    """Return true if label passes a criterion you set."""

img.label_threshold(filt=filt)

You can use additional arguments with names same as those properties supported in regionprops. For instance, you can label regions only satisfies proper area and length using following filter function.

def filt(img, lbl, area, major_axis_length):
    proper_size = 10 < area < 60**2
    proper_shape = 20 < major_axis_length < 120
    return proper_size and proper_shape

img.label_threshold(filt=filt)