import numpy as np 
from PIL import Image
from PIL import ImageFilter
from PIL import ImageEnhance

# Open images 
img = Image.open('./code.png')
img

# Filter
# There are many filters, like averaging and maximgum filter. 
# Each one is used for different purpose.
# Here, we use a median fileter to keep the edge and the details of the imgage
img = img.filter(ImageFilter.MedianFilter(5)) 
img

# Contrast Change
# Change paramters 'factor' to change the contrast
# We can increase the contrast to highlight characters in compare to other noise 
img = ImageEnhance.Contrast(img).enhance(1.3)#enhance()
img

# Scale change
# We do not care abhttps://mru4913.github.io/Night/posts/img/20200113/out the color 
# change it to grey scale 
img = img.convert('L')
img

# However as you can seewe still have some unnecessary 
# lines in the background 
# we need to find a way to remove those lines 
# this approach has to be carefully used b
def remove_pixel(img, vrange=(150,220)):
    pixdata = img.load()
    w, h = img.size
    for j in range(h):
        for i in range(w):
            if pixdata[i,j] > vrange[0] and  pixdata[i,j] < vrange[1]:
                pixdata[i,j] = 255

    return img

img = remove_pixel(img)
img

def denoise(im):
    pixdata = im.load()
    w, h = im.size
    for j in range(1,h-1):
        for i in range(1,w-1):
            count=0
            if pixdata[i,j-1] > 245:
                count = count+1
            if pixdata[i,j+1] > 245:
                count = count+1
            if pixdata[i+1,j] > 245:
                count = count+1
            if pixdata[i-1,j] > 245:
                count = count+1
            if count > 2:
                pixdata[i,j] = 255
    return im

# Binarize
# Now, we can make the characters clearer to see by binarizing.
def binarizing(img, threshold):
    pixdata = img.load()
    w, h = img.size
    for j in range(h):
        for i in range(w):
            if pixdata[i,j] < threshold:
                pixdata[i,j] = 0
            else:
                pixdata[i,j] = 255
    return img

img = binarizing(img, 200)
img

# Crop
# Then we may need to crop imgages
n = 4
w, h = img.size
imgg_w = w/n
for i in range(n):
    img.crop((0 + imgg_w * i, 0, imgg_w * (i+1), h)).show()

# Numpy to PIL image 
Image.fromarray(np.random.randint(0,255,(30,30)).astype('uint8'))

# PIl image to array
np.array(img)