SVD Compression

"""
Created on Tue Oct 27 12:31:49 2020.

@author: manoj
"""
import numpy as np
import matplotlib.pyplot as plt
import scipy.linalg as scipy_linalg
import skimage.io


def compress_svd(image, bw, color):
    U, s, Vt = scipy_linalg.svd(image, full_matrices=False)
    inspect_svalues(s, color)
    recon_image = U[:, 0:bw] @ scipy_linalg.diagsvd(s[0:bw], bw, bw) @ Vt[0:bw, :]
    return recon_image


def inspect_svalues(svals, color):
    plt.plot(svals)
    plt.xticks(list(range(0, 501, 40)))
    plt.xlabel("Index of Singular Value")
    plt.ylabel("Magnitude of Singular Value")
    plt.title("Singular Values of {0} Channel".format(color))
    plt.show()


def see_imgs(img1, img2, rank, color='gray'):
    fig, axs = plt.subplots(1, 2)
    axs[0].imshow(img1, cmap=color)
    axs[0].set_title("Original image")
    axs[1].imshow(img2, cmap=color)
    axs[1].set_title("Compressed image to rank = {0}".format(rank))
    plt.suptitle("Image compression using Singular Value Decomposition", y=0.82)
    plt.show()


eff_rank = 40

img = skimage.io.imread(fname="./img_for_q3.bmp")
olayers = [img[:, :, i].astype(np.double) for i in range(3)]
clayers = [compress_svd(olayers[i], eff_rank, c).astype(np.uint8) for i, c in enumerate(["Red", "Green", "Blue"])]
re_img = np.dstack(clayers)

see_imgs(img, re_img, eff_rank)
[see_imgs(olayers[i], clayers[i], eff_rank, color=cl) for i, cl in enumerate(['Reds_r', 'Greens_r', 'Blues_r'])]