How can i convert numpy array to pil image in memory?

Question

I have heatmap which is normalized 2d numpy array

When i plot it with matplotlib:

axes_img = plt.imshow(255 * normalized_heat_map, alpha=alpha, cmap=cmap)
plt.show()

However when i convert it to PIL image and then plot again

heatmap_image = Image.fromarray(np.uint8(axes_img.get_cmap()(axes_img.get_array()) * 255))
plt.imshow(np.asarray(heatmap_image))
plt.show()

How can I get image that is same to one from Matplotlib one without saving to file ?

Answer 1

The function imshow adjust the limits of the color scale to the minimum and maximum of the data you are plotting. If you keep this in mind, you realize you have to normalize the image array before converting it in an 8bit array. This can be done as follow:

scaled_img = (axes_img.get_array()-axes_img.get_clim()[0])/(axes_img.get_clim()[1]-axes_img.get_clim()[0])
heatmap_image = Image.fromarray(np.uint8(axes_img.get_cmap()(scaled_img) * 255))