main.py

import numpy as np
from numpy import linalg
import cv2
from math import sqrt
from math import atan
from math import pow
import glob
from PIL import Image
import os

for img in glob.glob("Original/*.jpg"):

    #getting image file name
    filename=os.path.basename(img)

    #reading the image
    a = cv2.imread(img)

    #making 3 grayscale images 
    b = cv2.cvtColor(a, cv2.COLOR_BGR2GRAY)
    c = cv2.cvtColor(a, cv2.COLOR_BGR2GRAY)
    d = cv2.cvtColor(a, cv2.COLOR_BGR2GRAY)

    #k-value range [0.1-0.5] can be changed depending on image
    k=0.2

    #r-value can be changed depending on image
    r=128

    #neighbouring window size can be changed depending on image
    window=35

    #otsu thresholding
    ret2,th= cv2.threshold(b,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)

    h,w=b.shape

    th=0
    value=[]
    threshold=[]
    m=0
    var=0
    std=0
    b,c=cv2.integral2(b)

    h,w=c.shape

    #Sauvola Method using integrl images
    for i in range(1, w , 1):
        for j in range(1, h , 1):
            if(j>(h-int(window/2)) and i>(w-int(window/2))):
                m = (b[j, i] + b[j - int(window/2), i - int(window/2)] - b[
                    j , i - int(window/2)] - b[j - int(window/2), i]) / (window * window)
                s = (c[j, i] + c[j - int(window/2), i - int(window/2)] - c[
                    j, i - int(window/2)] - c[j - int(window/2), i]) / (window * window)
            elif (i > (w - int(window/2)) and j < (h - int(window/2))):
                m = (b[j + int(window/2), i] + b[j - int(window/2), i - int(window/2)] - b[
                    j + int(window/2), i - int(window/2)] - b[j - int(window/2), i]) / (window * window)
                s = (c[j + int(window/2), i] + c[j - int(window/2), i - int(window/2)] - c[
                    j + int(window/2), i - int(window/2)] - c[j - int(window/2), i]) / (window * window)
            elif(j>(h-int(window/2)) and i<(w-int(window/2))):
                m = ( b[j,i+int(window/2)]+b[j-int(window/2),i-int(window/2)]-b[j,i-int(window/2)]-b[j-int(window/2),i+int(window/2)])/(window*window)
                s = (  c[j,i+int(window/2)]+c[j-int(window/2),i-int(window/2)]-c[j,i-int(window/2)]-c[j-int(window/2),i+int(window/2)])/(window*window)
            elif(j<(h-int(window/2) ) and i<(w-int(window/2))):
                m=(b[j+int(window/2),i+int(window/2)]+b[j-int(window/2),i-int(window/2)]-b[j+int(window/2),i-int(window/2)]-b[j-int(window/2),i+int(window/2)])/(window*window)
                s=(c[j+int(window/2),i+int(window/2)]+c[j-int(window/2),i-int(window/2)]-c[j+int(window/2),i-int(window/2)]-c[j-int(window/2),i+int(window/2)])/(window*window)
            var = ((s)- (pow((m), 2)))/(window*window)
            std = sqrt(abs(var))
            T = m * (1 + (k * ((std / r) - 1)))
            threshold.append(T)




    h1,w1=d.shape
    
    #setting pixel values based on computed threshold
    for i in range(0, w1, 1):
        for j in range(0, h1, 1):

            if (d[j, i] <= threshold[th]):
                d[j, i] = 0
            else:
                d[j, i] = 255
            th=th+1
            
    for i in range(0, int(window/2), 1):
        for j in range(0, h1, 1):
            d[j,i]=255
    for i in range(0, w1, 1):
        for j in range(0, int(window/2), 1):
            d[j,i]=255
            
    cv2.imwrite("Our_Output/"+str(filename),d)