以CSV格式保存阵列3D

现在还不清楚为什么要使用pandas数据帧来存储图像。我认为这让事情变得不必要的复杂。您可以直接以二进制格式存储numpy数组,并在以后的某个时刻再次加载它。

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

#create an image
imar = np.array([[[1.,0.,1.],[0.,0.,1.],[1.,1.,0.]],
                 [[0.,1.,1.],[0.,1.,1.],[1.,2.,3.]],
                 [[0.,0.,1.],[1.,1.,1.],[1.,0.,0.]]]).transpose()
plt.imsave('pic.png', imar)

#create dataframe

df = pd.DataFrame([[0,""]], columns=["Feature1","Feature2"])

# read the image
im = plt.imread('pic.png')

plt.imshow(im)
plt.show()

#save the image array to binary file
np.save('mypic.npy', im)
# store name of image in dataframe
df.iloc[0,1] = 'mypic.npy'
#save dataframe
df.to_csv("mydf.csv")
del df

#read dataframe from csv
df = pd.read_csv("mydf.csv")
# load the image from binary file, given the path from the Dataframe
new_im= np.load(df["Feature2"][0])
# show the loaded image
plt.imshow(new_im)
plt.show()

opencv ,为什么不使用 cv2.imwrite ,它与您的 cv2.imshow

开放式CV 图像正在使用 numpy.save 自从 开放式CV np.array .

如果你必须把它保存在 csv

          width . height .  0  .  1  . ...
image_1 .   w1  .   h1   .  f0 .  f1 . ...
image_2 .   w2  .   h2   .  f0 .  f1 . ...

其中每行包含图像的宽度和高度,后跟 image.flatten() . 然后可以将脚本修改为

new_dt=pd.read_csv('mypic2.csv')

for x in range(len(dt.index)):
    # recover image's dimension
    width = new_dt.iloc[x,0]
    height = new_dt.iloc[x,1]

    # load image and reshape to original dimension
    images = new_dt.iloc[x,2:].reshape(height, width, -1)
    print(imagess)
    cv2.imshow('imagenew', imagess)