如何在Python中使用步骤转换将图像正确显示为3D绘图？

Question

I am trying to visualize differences between images in 3D, in order to more easily differentiate between positive and negative differences.

I have succeeded with a basic plot of an image, however, between the values matplotlib is interpolating values. I need these to be step changes between pixels.

I am often testing with very low-resolution images, for example, 16 by 16, so the interpolation has a large effect.

Numpy file of 16 by 16 image: https://wetransfer.com/downloads/c916f76e0d86a61c00c2ed4cfe4ae97520190210192200/60d87c

One way to solve this would be to repeat the values however, this seems very inefficient and requires cleaning up the ticks after.

Code to generate above image:

import matplotlib.pyplot as plt
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

SubIm = np.load("Subtract_Image.npy")

def ImPlot2D3D(img, cmap=plt.cm.jet):

    Z = img[::1, ::1]

    fig = plt.figure(figsize=(14, 7))

    # 2D Plot
    ax1 = fig.add_subplot(1, 2, 1)
    im = ax1.imshow(Z, cmap=cmap)
    ax1.set_title('2D')
    ax1.grid(False)

    # 3D Plot
    ax2 = fig.add_subplot(1, 2, 2, projection='3d')
    X, Y = np.mgrid[:Z.shape[0], :Z.shape[1]]
    ax2.plot_surface(X, Y, Z, cmap=cmap)
    ax2.set_title('3D')

    plt.show()


ImPlot2D3D(SubIm)

I've looked into 3D bar charts but they all use binning schemes and I can't make it work for an image.

Answer 1

Eventually managed to answer my own question.

A brute force method to solve this is to repeat the values in the array, hence making the interpolation between values that 'matplotlib' does, less impactful and better approximating a step change. This can be achieved using numpy.repeat . As this is a 3D array we must iterate over one axis than the other. Otherwise, the array will be flattened repeated and this flat array returned.

Result:

def ImPlot2D3D(img, cmap=plt.cm.jet, step=False, ratio=10):

    if step:
        img = (img.repeat(ratio, axis=0)).repeat(ratio, axis=1)

    Z = img[::1, ::1]

    fig = plt.figure(figsize=(14, 7))

    # 2D Plot
    ax1 = fig.add_subplot(1, 2, 1)
    im = ax1.imshow(Z, cmap=cmap)
    ax1.set_title('2D')
    ax1.grid(False)

    # 3D Plot
    ax2 = fig.add_subplot(1, 2, 2, projection='3d')
    X, Y = np.mgrid[:Z.shape[0], :Z.shape[1]]
    ax2.plot_surface(X, Y, Z, cmap=cmap)
    ax2.set_title('3D')

    # Scale the ticks back down to original values
    if step:
        ticks_x = ticker.FuncFormatter(lambda x, pos: '{0:g}'.format(x / ratio))
        ticks_y = ticker.FuncFormatter(lambda y, pos: '{0:g}'.format(y / ratio))
        ax1.xaxis.set_major_formatter(ticks_x)
        ax1.yaxis.set_major_formatter(ticks_y)
        ax2.xaxis.set_major_formatter(ticks_x)
        ax2.yaxis.set_major_formatter(ticks_y)

    plt.show()

import matplotlib.ticker as ticker
import matplotlib.pyplot as plt
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

SubIm = np.load("Subtract_Image.npy")
ImPlot2D3D(SubIm, step=True)