将带有图像数据的 numpy 数组转换为长格式

Question

我想将图像转换为“长格式”并返回。我现在的代码是import numpy as npimport pandas as pdimg = mpimg.imread('/path/image.png')image = []for i in range(img.shape[0]):    for j in range(img.shape[1]):        image.append([i, j] + img[i, j].tolist())image = np.array(image)但我确信有一个更高效的，但奇怪的是在互联网上找不到任何东西。什么是更快的解决方案？上面的代码产生了正确的结果。我的图像是彩色的，img.shape三元组也是如此。

Answer 1

这是一种方法 -

m,n = img.shape[:2]

r,c = np.mgrid[:m,:n]

out = np.column_stack((r.ravel(), c.ravel(), img.reshape(-1,img.shape[2])))

替代获得r,c：

r,c = np.indices(img.shape[:2])

另一个带有数组分配的 -

m,n,r = img.shape

out = np.empty((m,n,2+r), dtype=img.dtype)

out[:,:,0] = np.arange(m)[:,None]

out[:,:,1] = np.arange(n)

out[:,:,2:] = img

out = out.reshape(m*n,-1)

Answer 2

这是一种使用np.indices，转置以匹配您的循环并重新整形以获得二维数组的方法。

ix = np.transpose(np.indices(img.shape[:2]), (1,2,0))

image = np.concatenate((ix, img), axis=2).reshape(-1, image.shape[2] + 2)

Answer 3

只是为了获得时间：

import matplotlib.image as mpimg

import numpy as np

import pandas as pd

import time

t0 = time.time()

image_orig = []

for i in range(img.shape[0]):

    for j in range(img.shape[1]):

        image_orig.append([i, j] + img[i, j].tolist())

image_orig = np.array(image_orig)

print(time.time() - t0)

t0 = time.time()

ix = np.transpose(np.indices(img.shape[:2]), (1,2,0))

image = np.concatenate((ix, img), axis=2).reshape(-1, img.shape[2] + 2)

print(time.time() - t0)

t0 = time.time()

m,n = img.shape[:2]

r,c = np.indices(img.shape[:2])

out = np.column_stack((r.ravel(), c.ravel(), img.reshape(-1,img.shape[2])))

print(time.time() - t0)

t0 = time.time()

m,n,r = img.shape

out = np.empty((m,n,2+r), dtype=img.dtype)

out[:,:,0] = np.arange(m)[:,None]

out[:,:,1] = np.arange(n)

out[:,:,2:] = img

out = out.reshape(m*n,-1)

print(time.time() - t0)

0.17211008071899414

0.001434326171875

0.0013523101806640625

0.0008423328399658203

最后一个变体似乎是最快的。