Python：获取数组所有掩码的最快方法

Question

Is there a faster way than looping through all components of a 2d array to get all possible masks in a specific range like:

import numpy as np
numOfLabels = 80
array2D = np.random.choice(255,(512,512))
for i in range(1,numOfLabels):
    mask = array2D==i

Maybe something with broadcasting and creating a 3d array with all masks

EDIT:

Thanks for the already written answers. For better explanation. What im trying to do:

I have an 2d label matrix with components. The components are labeled with numbers, lets say 1 to 80. I have two images. Now i want to calculate the mean,max,min of those two images for all 80 compoents each. Maybe I had the totaly wrong idea to do that.

EDIT2:

According to the comment I found a way with the following code:

from scipy import ndimage
import numpy as np

def calculateMeanMaxMin(val):
    return np.array([np.mean(val),np.max(val),np.min(val)])

def getTheStatsForComponents(array1,array2):
    ret, thresholded= cv2.threshold(array2, 50, 255, cv2.THRESH_BINARY)
    thresholded= thresholded.astype(np.uint8)
    numLabels, labels, stats, centroids = cv2.connectedComponentsWithStats(thresholded, 8, cv2.CV_8UC1)
    allComponentStats=[]
    meanmaxminArray2 = ndimage.labeled_comprehension(array2, labels, np.arange(1, numLabels+1), calculateMeanMaxMin, np.ndarray, 0)
    meanmaxminArray1 = ndimage.labeled_comprehension(array1, labels, np.arange(1, numLabels+1), calculateMeanMaxMin, np.ndarray, 0)
    for position, label in enumerate(range(1, numLabels)):
        currentLabel = np.uint8(labels== label)
        _, contour, _ = cv2.findContours(currentLabel, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
        (side1,side2)=cv2.minAreaRect(contour[0])[1]
        componentStat = stats[label]
        allstats = {'position':centroids[label,:],'area':componentStat[4],'height':componentStat[3],
                              'width':componentStat[2],'meanArray1':meanmaxminArray1[position][0],'maxArray1':meanmaxminArray1[position][1],
                              'minArray1':meanmaxminArray1[position][2],'meanArray2':meanmaxminArray2[position][0],'maxArray2':meanmaxminArray2[position][1],
                              'minArray2':meanmaxminArray2[position][2]}
        allComponentStats.append(allstats)
    return allComponentStats

But I wonder if there is a faster way to get all Stats for all components written to a dictionary , because i dont know if ndimage.measurements.labeled_comprehension is fast, and if i somehow can avoid this loop?

Answer 1

You can use np.put_along_axis to create all masks at once. All non labeled points go to plane zero:

all_masks = np.zeros((numOfLabels, *array2D.shape), bool)
np.put_along_axis(all_masks, array2D*(array2D<numOfLabels)[None], True, axis=0)

# check
for i in range(1,numOfLabels):
    assert (all_masks[i] == (array2D==i)).all()

Answer 2

This creates the 3-D array of masks, using broadcasting:

mask = array2D == np.arange(1, numOfLabels)[:, None, None]

Eg

In [26]: numOfLabels = 80                                                                                          

In [27]: array2D = np.random.choice(255, (512, 512))                                                               

In [28]: mask = array2D == np.arange(1, numOfLabels)[:, None, None]                                                

In [29]: mask.shape                                                                                                
Out[29]: (79, 512, 512)

In [30]: array2D[:10, :13]                                                                                         
Out[30]: 
array([[111, 161, 109, 235, 193, 224,  63,  78, 106, 245, 140,  64,  28],
       [245, 239, 225,  31, 239, 212, 137,  17, 130, 185, 110,  70,  55],
       [251,  65, 114, 199, 229, 241,  97,  84,  13,  10,   4, 114, 185],
       [217, 252, 205,  94,  93, 202,  99,  91,  65,  34,  86,  84, 233],
       [115,  51, 217, 105, 187,  61, 203,  30, 178, 178, 183, 193, 231],
       [ 75, 195,  21, 143, 180,  32,  38, 224, 188,  85,  80, 250,   4],
       [163, 174,  35,  49, 202, 110, 223, 209,  80,  38, 127,  31, 208],
       [137, 133,  41,  30, 193, 187, 182, 171,  72, 151, 180, 152,  14],
       [145, 108, 112, 254,  92,  87,  45, 173,  45,  28, 189, 248,  48],
       [147, 222,  37,  31, 198,  69,   8, 136,  85, 162, 225, 203,  68]])

Note: I copied the range of labels from your code, where you start the range with 1 . That means the mask for the label 4 will be at index 3 :

In [31]: mask[3, :10, :13].astype(int)  # Display bools as ints for compact output                                                                               
Out[31]: 
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])

Answer 3

array2D==i give you the position where array2D == i, although I do not understand what you are trying to do, you can modify the code by

import numpy as np
numOfLabels = 80
array2D = np.random.choice(255,(512,512))
mask = np.zeros((512,512,numOfLabels))
for i in range(numOfLabels):
    mask[array2D==i, i] = 1