将图像分割成 m*n 块重叠

Question

Currently, I can split the image but having trouble doing the overlapping. The overlapping logic is shown in the image below. Ideally, overlapping can be set using the overlap variable which ranges from 0 to 1.

import numpy as np
from PIL import Image


rows = 2
columns = 3
overlap = 0.3

def split_image_to_tiles(image):
    tiles = np.array(image)
    tiles = np.array_split(tiles, rows, axis=1)
    tiles = [np.array_split(t, columns, axis=0) for t in tiles]
    tiles = np.array(tiles, dtype='uint8')
    return tiles

def overlap_tiles(tiles):
    for i in range(rows):
        for j in range(columns):
            tile = Image.fromarray(tiles[i][j])
            #overlapping logic
            
            tiles[i][j] = np.array(tile)
    return tiles

if __name__ == '__main__':
    image = Image.open('test')
    tiles = split_image_to_tiles(image)
    overlap_tiles(tiles)
    for i in range(rows):
        for j in range(columns):
            tile = Image.fromarray(tiles[i][j])
            tile.show()

Answer 1

You could use view_as_windows in the following way:

from skimage.util.shape import view_as_windows

def split_to_overlapping_windows(image, window_size, stride):    
    image_windows = view_as_windows(image,
                                    window_shape=(window_size, window_size, image.shape[-1]),
                                    step=stride).squeeze()

    window_layout_shape = image_windows.shape[0:2]

    image_windows = image_windows.reshape(-1, *image_windows.shape[2:])

    return image_windows, window_layout_shape

• Quick search gave me the following question: Splitting Very large images into overlapping boxes/blocks/tiles/sections, python, OpenCV that solves the problem better since view_as_windows is very memory intensive. (read the note in the doc)