CODEWITHSUNDEEP
pythonopencvimage-processingcomputer-vision
user19249111
user19249111

Reputation:

How to transfer the colors of an image to another?

How do I get the image color tones of an image and set it to another one?

I have these two images and want to make Ashley Benson in Mona Lisa's image colors.

enter image description here enter image description here

Upvotes: 0

Views: 1064

Answers (1)

Maxim Lyuzin
Maxim Lyuzin

Reputation: 397

I don't think there could be a "filter" to do it.

Using classical computer vision you could make Fast Fourier Transform of each image, and than replace low-frequency components of Ashley Benson image with Mona Lisa's. But in this case you can only change the color domain of the image. Code example here:

import cv2
import numpy as np
from matplotlib import pyplot as plt

lisa = cv2.imread(r"path/to/monalisa")
ashley = cv2.imread(r"path/to/ashley")

def domain_adoptation(src, trg, freq):

    """
    Parameters: 
    src - source image, which style has to be changed
    trg - target image, which low-frequency domain will be adopted
    freq - number of frequencies to be used

    Returns:
    result - np.array based on srs image (shape and high frequencies) 
         with low frequencies of the target image
    """

    result = np.zeros((src.shape[0],src.shape[1],src.shape[2]))

    for i in range(src.shape[2]):
        trg_fft = np.fft.fft2(trg[:,:,i])
        src_fft = np.fft.fft2(src[:,:,i])

        trg_fft_shift = np.fft.fftshift(trg_fft)
        src_fft_shift = np.fft.fftshift(src_fft)

        src_fft_shift[src.shape[0]//2-freq:src.shape[0]//2+freq,
                         src.shape[1]//2-freq:src.shape[1]//2+freq] = \
            trg_fft_shift[trg.shape[0]//2-freq:trg.shape[0]//2+freq,
                           trg.shape[1]//2-freq:trg.shape[1]//2+freq]

        src_ifft_shift = np.fft.ifftshift(src_fft_shift)

        result[:,:,i] = np.fft.ifft2(src_ifft_shift)
        result[:,:,i] = np.abs(result[:,:,i])

    result = np.float32(result)
    result = cv2.cvtColor(result, cv2.COLOR_BGR2RGB)
    result = cv2.normalize(result,None,0,1,cv2.NORM_MINMAX)

    return result

image = domain_adoptation(src=ashley,trg=lisa,freq=1)
plt.imshow(a)

And there's a GIF:

Ashley to Lisa

If you want to have better results, you can take a look at the quite old deep learning method called "Fast neural style". To do this with Mona Lisa's image, you need to train your own model using the examples above. You can check pretrained models in this colab. Trained models (for each style its own model) give these results:

candy mosaic

Of course there are many modern state-of-the-art approaches of style transfer, see here.

Upvotes: 3

Related Questions