Store a numpy.ndarray as an image, and then save t

问题:

I am currently looking for at in which i can store a numpy.ndarray as an image, then save that image, and extract the pixel values of the image into a numpy.ndarray. The dimension of the numpy.ndarray with the pixel values should be the same as the numpy.ndarray what was used to create the plot.

I tried something like this:

def make_plot_store_data(name, data):
    plt.figure()
    librosa.display.specshow(data.T,sr=16000,x_axis='frames',y_axis='mel',hop_length=160,cmap=cm.jet)
    plt.savefig(name+".png")
    plt.close()

    convert = plt.get_cmap(cm.jet)
    numpy_output_interweawed = convert(data.T)

The first image looks like this: The second image looks like this:

Why is this so messed up?

回答1:

Here's one approach that takes a 512x512 ndarray, displays it as an image, stores it as an image object, saves an image file, and generates a normalized pixel array of the same shape as the original.

import numpy as np

# sample numpy array of an image
from skimage import data
camera = data.camera()
print(camera.shape) # (512, 512)

# array sample
print(camera[0:5, 0:5])

[[156 157 160 159 158]
 [156 157 159 158 158]
 [158 157 156 156 157]
 [160 157 154 154 156]
 [158 157 156 156 157]]

# display numpy array as image, save as object
img = plt.imshow(camera)

# save image to file
plt.savefig('camera.png')

# normalize img object pixel values between 0 and 1
normed_pixels = img.norm(camera)

# normed_pixels array has same shape as original 
print(normed_pixels.shape) # (512, 512)

# sample data from normed_pixels numpy array 
print(normed_pixels[0:5,0:5])

[[ 0.61176473  0.6156863   0.627451    0.62352943  0.61960787]
 [ 0.61176473  0.6156863   0.62352943  0.61960787  0.61960787]
 [ 0.61960787  0.6156863   0.61176473  0.61176473  0.6156863 ]
 [ 0.627451    0.6156863   0.60392159  0.60392159  0.61176473]
 [ 0.61960787  0.6156863   0.61176473  0.61176473  0.6156863 ]]

You might consider looking into the skimage module, in addition to standard pyplot methods. There are a bunch of image manipulation methods there, and they're all built to play nice with numpy. Hope that helps.