如何用Python将图像转换为NumPy数组并保存为CSV文件

让我们看看如何在Python中把一张图片转换为NumPy数组，然后把这个数组保存到CSV文件中？首先，我们将学习如何将图像转换为Numpy ndarray。有很多方法可以将图像转换为ndarray，其中有几种方法。

方法1：使用PIL和NumPy库。

我们将使用PIL.Image.open()和numpy.asarray()。

示例:

# import required libraries
from PIL import Image
import numpy as gfg
 
# read an image
img = Image.open('geeksforgeeks.jpg')
 
# convert image object into array
imageToMatrice = gfg.asarray(img)
 
# printing shape of image
print(imageToMatrice.shape)

输出:

(251, 335, 3)

方法2：使用Matplotlib库。

我们将使用matplotlib.image.imread()方法。

示例:

# import library
from matplotlib.image import imread
 
# read an image
imageToMatrice  = imread('geeksforgeeks.jpg')
 
# show shape of the image
print(imageToMatrice.shape)

输出:

(251, 335, 3)

现在，imageToMatrice变量包含了从给定图像转换后得到的ndarray。

获得的矩阵的尺寸由图像中存在多少个通道决定。

• 对于一个黑白或灰度图像。只有一个通道存在，因此，矩阵的形状将是（n，n），其中n代表图像的尺寸（像素），矩阵内的值范围为0至255。
• 对于彩色或RGB图像。它将呈现3个通道的张量，因此矩阵的形状是（n, n,3）。每个通道是一个(n, n)矩阵，每个条目分别代表图像内部实际位置的红、绿或蓝的水平。

我们将使用两种方法来做同样的事情，第一种方法使用numpy库，第二种方法使用pandas库。

注意：我们只能在文件中保存一维或二维矩阵，因此，在灰度或黑白图像中不会有问题，因为它是一个二维矩阵，但我们需要确保这对彩色或RGB图像有效，因为它是一个三维矩阵。

方法1：使用NumPy库。

我们将使用numpy.savetxt（）和numpy.loadtxt（）。

示例:

# import required libraries
import numpy as gfg
import matplotlib.image as img
 
# read an image
imageMat = img.imread('gfg.jpg')
print("Image shape:", imageMat.shape)
 
# if image is colored (RGB)
if(imageMat.shape[2] == 3):
   
  # reshape it from 3D matrice to 2D matrice
  imageMat_reshape = imageMat.reshape(imageMat.shape[0],
                                      -1)
  print("Reshaping to 2D array:",
        imageMat_reshape.shape)
 
# if image is grayscale
else:
  # remain as it is
  imageMat_reshape = imageMat
     
# saving matrice to .csv file
gfg.savetxt('geek.csv',
            imageMat_reshape)
 
# retrieving matrice from the .csv file
loaded_2D_mat = gfg.loadtxt('geek.csv')
 
# reshaping it to 3D matrice
loaded_mat = loaded_2D_mat.reshape(loaded_2D_mat.shape[0],
                                   loaded_2D_mat.shape[1] // imageMat.shape[2],
                                   imageMat.shape[2])
 
print("Image shape of loaded Image:",
      loaded_mat.shape)
 
# check if both matrice have same shape or not
if((imageMat == loaded_mat).all()):
   
  print("\n\nYes",
        "The loaded matrice from CSV file is same as original image matrice")

输出:

Image shape: (251, 335, 3)
Reshaping to 2D array:(251, 1005)
Image shape of loaded Image:（251, 335, 3）。
Yes The loaded matrice from CSV file is same as original image matrice

方法2：使用Pandas库。

我们将使用pandas.Dataframe( )和pandas.Dataframe() .to_csv()方法。

# import required libraries
import numpy as gfg
import matplotlib.image as img
import pandas as pd
 
# read an image
imageMat = img.imread('gfg.jpg')
print("Image shape:",
      imageMat.shape)
 
# if image is colored (RGB)
if(imageMat.shape[2] == 3):
   
  # reshape it from 3D matrice to 2D matrice
  imageMat_reshape = imageMat.reshape(imageMat.shape[0],
                                      -1)
  print("Reshaping to 2D array:",
        imageMat_reshape.shape)
 
# if image is grayscale
else:
  # remain as it is
  imageMat_reshape = imageMat
     
# converting it to dataframe.
mat_df = pd.DataFrame(imageMat_reshape)
 
# exporting dataframe to CSV file.
mat_df.to_csv('gfgfile.csv',
              header = None,
              index = None)
 
# retrieving dataframe from CSV file
loaded_df = pd.read_csv('gfgfile.csv',
                        sep = ',',
                        header = None)
# getting matrice values.
loaded_2D_mat = loaded_df.values
 
# reshaping it to 3D matrice
loaded_mat = loaded_2D_mat.reshape(loaded_2D_mat.shape[0],
                                   loaded_2D_mat.shape[1] // imageMat.shape[2],
                                   imageMat.shape[2])
 
print("Image shape of loaded Image :",
      loaded_mat.shape)
 
# check if both matrice have same shape or not
if((imageMat == loaded_mat).all()):
  print("\n\nYes",
        "The loaded matrice from CSV file is same as original image matrice")

输出:

Image shape: (251, 335, 3)
Reshaping to 2D array: (251, 1005)
Image shape of loaded Image : (251, 335, 3)
Yes The loaded matrice from CSV file is same as original image matrice