将文件转换为灰度图像

您无需将数据转换为十六进制或二进制，只需将二进制数据（字节序列）转换为2D数组即可。

问题在于，没有任何一维数组可以重塑为二维数组。
例如，如果字节数是质数= N，则将获得1xN图像（丑陋的单行或列图像）。

以下示例假定图像尺寸必须为正方形，并根据需要使用填充来完成字节数：

import numpy as np
from math import sqrt, ceil
import cv2

#Input file name (random file I found in my folder).
input_file_name = 'test_cython.cp36-win_amd64.pyd';

#Read the whole file to data
with open(input_file_name, 'rb') as binary_file:        
    data = binary_file.read()

# Data length in bytes
data_len = len(data)

# d is a verctor of data_len bytes
d = np.frombuffer(data, dtype=np.uint8)

# Assume image shape should be close to square
sqrt_len = int(ceil(sqrt(data_len)))  # Compute square toot and round up

# Requiered length in bytes.
new_len = sqrt_len*sqrt_len

# Number of bytes to pad (need to add zeros to the end of d)
pad_len = new_len - data_len

# Pad d with zeros at the end.
# padded_d = np.pad(d, (0, pad_len))
padded_d = np.hstack((d, np.zeros(pad_len, np.uint8)))

# Reshape 1D array into 2D array with sqrt_len pad_len x sqrt_len (im is going to be a Grayscale image).
im = np.reshape(padded_d, (sqrt_len, sqrt_len))

# Save image
cv2.imwrite('im.png', im)

# Display image
cv2.imshow('im' ,im)
cv2.waitKey(0)
cv2.destroyAllWindows()

结果：

构建可用于还原原始文件的映像：

如果要获取图像并还原原始文件（字节相等，不带填充），则需要还原图像中的原始数据长度。
（您也可以恢复填充的长度）。

以下实现将原始数据长度存储在前8个像素中。

读取图像后，可以删除填充和存储的长度，并可以恢复原始文件。

这是一个“编码”和“解码”示例：

import numpy as np
from math import sqrt, ceil
import cv2
import struct

#Input file name
input_file_name = 'test_cython.cp36-win_amd64.pyd';

#Read the whole file to data
with open(input_file_name, 'rb') as binary_file:        
    data = binary_file.read()

# Data length in bytes
data_len = len(data)

# d is a verctor of data_len bytes
d = np.frombuffer(data, dtype=np.uint8)

data_len_as_bytes = np.frombuffer(struct.pack("Q", data_len), dtype=np.uint8) # Convert data_len to 8 bytes

data_len = data_len + len(data_len_as_bytes) #Update length to include the 8 bytes

# Set data_len as first 8 bytes of d
d = np.hstack((data_len_as_bytes, d))

# Assume image shape should be close to square
sqrt_len = int(ceil(sqrt(data_len)))  # Compute square toot and round up

# Requiered length in bytes.
new_len = sqrt_len*sqrt_len

# Number of bytes to pad (need to add zeros to the end of d)
pad_len = new_len - data_len

# Pad d with zeros at the end.
# padded_d = np.pad(d, (0, pad_len))
padded_d = np.hstack((d, np.zeros(pad_len, np.uint8)))

# Reshape 1D array into 2D array with sqrt_len pad_len x sqrt_len (im is going to be a Grayscale image).
im = np.reshape(padded_d, (sqrt_len, sqrt_len))

# Save image
cv2.imwrite('im.png', im)

# Display image
#cv2.imshow('im' ,im)
#cv2.waitKey(0)
#cv2.destroyAllWindows()


# Restore original data:
##################################
input_file_name = 'test.bin';  #Output file name

im = cv2.imread('im.png', cv2.IMREAD_GRAYSCALE)

# Convert 2D to 1D
padded_d = im.flatten()

# Get original length
data_len_as_bytes = padded_d[0:8]

orig_data_len = struct.unpack("Q", data_len_as_bytes.tobytes())

# Crop the original data bytes (without the padding).
data = padded_d[8:8+orig_data_len[0]]

#Write d whole file to binary file
with open(input_file_name, 'wb') as binary_file:
    binary_file.write(data)

现在，您可以将任何（小）文件作为映像上传到Stack Overflow，然后让其他人还原您的文件。