sarbina cute cute

main.py

from scipy import fftpack
import imageio
import numpy

# load in an image, convert to grayscale if needed
image = imageio.imread('ft/input.png', as_gray=True)

# take the fourier transform of the image
fft2 = fftpack.fftshift(fftpack.fft2(image))

# save FFT to a file. To help with visualization, we take
# the log of the magnitudes, and then stretch them so they
# fill the whole range of pixel values from 0 to 255.
imageio.imsave(
    'ft/fft.png',
    (numpy.log(abs(fft2)) * 255 / numpy.amax(numpy.log(abs(fft2)))).astype(
        numpy.uint8))

# At this point, fft2 is just a numpy array and you can
# modify it in order to modify the image in the frequency
# space. Here's a little example (that makes a nearly
# imperceptible change, but demonstrates what you can do.

# example remove high freq
print(fft2.shape)

# threshold = 80
middle = 103


def lowerIntenseRegion(fft2, start_x, width, start_y, height, ratio):
    for i in range(middle - start_y - height, middle - start_y):
        for j in range(middle - start_x - width, middle - start_x):
            fft2[i, j] = 10e-6 + 10e-6j

    for i in range(middle + start_y, middle + start_y + height):
        for j in range(middle + start_x, middle + start_x + width):
            fft2[i, j] = 10e-6 + 10e-6j


# remove noise
lowerIntenseRegion(fft2, 12, 12, 14, 10, 100)

lowerIntenseRegion(fft2, 80, 20, 20, 80, 200)

# bandfreq filter
# for i in range(206):
#     for j in range(206):
#         dist = (i - middle)**2 + (j - middle)**2
#         if dist < 50**2 and dist > 30**2:
#             fft2[i, j] = 10e-6 + 10e-6j

imageio.imsave(
    'ft/fft-2.png',
    (numpy.log(abs(fft2)) * 255 / numpy.amax(numpy.log(abs(fft2)))).astype(
        numpy.uint8))

# now take the inverse transform to convert back to an image
ifft2 = abs(fftpack.ifft2(fftpack.ifftshift(fft2)))

# and save the image
imageio.imsave('ft/fft-then-ifft.png', ifft2.astype(numpy.uint8))