8.7

'''
Učitati sliku lena.png i dodati joj AWGN varijanse 100 i srednje vrednosti 0.
Nad ovom zašumljenom slikom dodati impulsni so i biber šum gustine 5%.

Za potrebe uklanjanja ovog kombinovanog šuma potrebno je napisati funkciju
alphaTrimmedMeanFilt koja implementira alfa-trimovani usrednjivač dat formulom:
'''

from skimage import io
import plotly.express as px
import numpy as np

img = io.imread('lena.png')
fig = px.imshow(img, color_continuous_scale='gray')
fig.show()

rng = np.random.default_rng()

img_noisy = img + rng.standard_normal(size=img.shape)*np.sqrt(100) + 0
img_noisy[img_noisy>255] = 255
img_noisy[img_noisy<0] = 0

p = rng.uniform(size=img.shape)
p0 = 0.05
img_noisy[p<p0/2] = 0
img_noisy[(p0/2 <= p) & (p<p0)] = 255

fig = px.imshow(img_noisy, color_continuous_scale='gray')
fig.show()

import math

def alphaTrimmedMeanFilt(g, S_shape, alpha):
    f = np.zeros(g.shape)

    pad_rows = math.floor(S_shape[0]/2)
    pad_cols = math.floor(S_shape[1]/2)

    img_p = np.pad(g, ((pad_rows,),(pad_cols,)), mode='symmetric')

    for row in range(f.shape[0]):
        for col in range(f.shape[1]):
            region = img_p[row:row+S_shape[0], col:col+S_shape[1]]

            region = np.sort(region, axis = None)

            res = region[alpha//2:-alpha//2].mean()

            f[row,col] = res
    return f

# 3x3 region
S_shape = (3,3)

img_f = alphaTrimmedMeanFilt(img_noisy, S_shape, alpha=4)

fig = px.imshow(img_f, color_continuous_scale='gray')
fig.show()

psnr1 = 10*np.log10(255**2/((img - img_noisy)**2).mean())
psnr2 = 10*np.log10(255**2/((img - img_f)**2).mean())

print("PSNR za zasumljenu sliku je {}".format(psnr1))
print("PSNR za alfa-trimovani usrednjivac dimenzije {} je {}".format(S_shape, psnr2))