将大量图像拼接在一起的大多数pythonic方法

Question

我有一系列来自自动显微镜程序的图像，从96孔板获取图像（下图）。每个图像都命名为：

<Well name>_s<site number>.tif

96孔板的每个孔进一步细分为位点，位点以X×Y矩阵排列，并且数字逐行排列（下图）。

例如，将命名96孔板中左上孔的第（10,9）位点的图像 A01_s90.tif。

将图像拼接在一起的最Pythonic方法是什么？我目前正在使用OpenCV加载图像并在四个numpy.concatenate循环中调用for，但这看起来非常笨拙。

Answer 1

类似下面的代码？我假设它是问题的拼接，而不是文件名和井/站点索引之间的转换。

import numpy as np
import matplotlib.pyplot as plt

img_xsize, img_ysize = 20, 20 # size of single image
mx, my = 10, 10 # x/y grid of sites per well
nx, ny = 12, 8 # x/y grid of wells per plate
wgap = 20 # pixels gap between wells
stitched_x, stitched_y = (wgap+mx*img_xsize)*nx, (wgap+my*img_ysize)*ny

img_stitched = np.zeros((stitched_y, stitched_x), dtype=np.uint8)

def add_img(mxi, myi, nxi, nyi, img):
    assert img.shape == (img_ysize, img_xsize)
    xi = nxi*(mx*img_xsize + wgap) + mxi*img_xsize
    yi = nyi*(my*img_ysize + wgap) + myi*img_ysize
    img_stitched[yi:yi+img_ysize,xi:xi+img_xsize] = img

for nxi in range(nx):
    for nyi in range(ny):
        for mxi in range(mx):
            for myi in range(my):
                # ... get image data
                img = np.random.randint(0,255) * np.ones((img_ysize, img_xsize), dtype=np.uint8)
                add_img(mxi, myi, nxi, nyi, img)

plt.imshow(img_stitched)
plt.colorbar()
plt.show(block=False)
raw_input("Enter")

Answer 2

Han-Kwang Nienhuys的答案并没有消除困扰OP的四个嵌套for循环。

这是他的答案的修改版本，只有一个for循环：

import numpy as np
import matplotlib.pyplot as plt
from itertools import product

img_xsize, img_ysize = 20, 20 # size of single image
mx, my = 10, 10 # x/y grid of sites per well
nx, ny = 12, 8 # x/y grid of wells per plate
wgap = 20 # pixels gap between wells
stitched_x, stitched_y = (wgap+mx*img_xsize)*nx, (wgap+my*img_ysize)*ny

img_stitched = np.zeros((stitched_y, stitched_x), dtype=np.uint8)

def add_img(mxi, myi, nxi, nyi, img):
    assert img.shape == (img_ysize, img_xsize)
    xi = nxi*(mx*img_xsize + wgap) + mxi*img_xsize
    yi = nyi*(my*img_ysize + wgap) + myi*img_ysize
    img_stitched[yi:yi+img_ysize,xi:xi+img_xsize] = img

for nxi, nyi, mxi, myi in product(range(nx), range(ny), range(mx), range(my)):
    # ... get image data
    img = np.random.randint(0,255) * np.ones((img_ysize, img_xsize), dtype=np.uint8)
                add_img(mxi, myi, nxi, nyi, img)

plt.imshow(img_stitched)
plt.colorbar()
plt.show(block=False)
raw_input("Enter")