numpy 圆形掩码切口在二维数组的边缘被截断

Question

我有一个 2D numpy 数组，我想屏蔽几个特定元素周围的圆形区域。这对于不靠近边缘的元素很容易。但是当磁盘在靠近数组边缘的位置被截断时，掩码操作会引发异常。

屏蔽这些边缘源的好方法是什么？定期填充..？谢谢！

这里有一些代码：

import numpy as np

x_count = 20000
y_count = 20000

a = np.arange(y_count * x_count).reshape(y_count, x_count) # data
master_mask=np.zeros(a.shape).astype(bool)

coords=([10,50],[500,400])
nsrc=len(coords)
mask_radius=100 # 10, 20, etc.

for isrc in range(nsrc):
    xc=coords[isrc][0]; yc=coords[isrc][1]
    x,y = np.ogrid[-mask_radius:mask_radius+1,-mask_radius:mask_radius+1]
    minimask = x*x + y*y <= mask_radius*mask_radius
    box=master_mask[xc-mask_radius:xc+mask_radius+1,yc-mask_radius:yc+mask_radius+1]
    assert(minimask.shape==box.shape), 'This assertion should fail for discs touching the edge'
    master_mask[xc-mask_radius:xc+mask_radius+1,yc-mask_radius:yc+mask_radius+1] += minimask

print master_mask

Answer 1

这是一个可能的（行人）答案，虽然我希望有一种更 numpy-esque 的方式。

import numpy as np

x_count = 20000
y_count = 20000

a = np.arange(y_count * x_count).reshape(y_count, x_count) # data
master_mask=np.zeros(a.shape).astype(bool)

coords=([10,50],[500,400])
nsrc=len(coords)
mask_radius=100 # 10, 20, etc.

for isrc in range(nsrc):
    xc=coords[isrc][0]; yc=coords[isrc][1]
    # Truncate the postage-stamp mask here
    xmin=max(xc-mask_radius,0); xmax=min(xc+mask_radius+1,x_count)
    ymin=max(yc-mask_radius,0); ymax=min(yc+mask_radius+1,y_count)
    x,y = np.ogrid[xmin:xmax,ymin:ymax]

    minimask = x*x + y*y <= mask_radius*mask_radius
    box=master_mask[xmin:xmax,ymin:ymax]
    #assert(minimask.shape==box.shape), 'This assertion should fail for discs near the edge'
    master_mask[xmin:xmax,ymin:ymax] += minimask

print master_mask