Variable Area Threshold For Identifying Objects - Python

I have an array which contains information of the size and location a series of shapes: where the array is zero, there are no shapes, where the array is not zero there is a shape.

Solution 1:

You can use scipy.ndimage.label to find the connected non-zero regions in your array, then use scipy.ndimage.sum to find the area of each region:

from scipy import ndimage

labels, nshapes = ndimage.label(a)
areas = ndimage.sum(a, labels=labels, index=range(1, nshapes))

idx = np.argmax(areas)
biggest_shape = labels == (idx + 1)

In your example there happen to be two 'shapes' with the same area:

from matplotlib import pyplot as plt

fig, (ax1, ax2, ax3) = plt.subplots(1, 3)

ax1.imshow(a, cmap=plt.cm.jet)
ax2.imshow(labels, cmap=plt.cm.jet)
ax3.imshow(biggest_shape, cmap=plt.cm.jet)

Update

The structure argument passed to scipy.ndimage.label determines which neighbouring elements are considered to be 'connected' (see the docs linked above). If you want diagonally adjacent elements to be considered as connected, you can pass a 3x3 array of ones:

labels, nshapes = ndimage.label(a, structure=np.ones((3, 3)))