Line removal in image captcha using Python

Question

I had used this link - How to remove line from captcha completely and edited the code provided to remove lines from a dummy captcha that I have given below

lineRemoval.py

from PIL import Image,ImageFilter
from scipy.misc import toimage
from operator import itemgetter
from skimage import measure
import numpy as np
import heapq
import cv2
import matplotlib.pyplot as plt
from scipy.ndimage.filters import median_filter



#----------------------------------------------------------------

class preprocessing:
    def pre_proc_image(self,img):
        img_removed_noise=self.apply_median_filter(img)
        #img_removed_noise=self.remove_noise(img)
        p1,p2,LL=self.get_line_position(img_removed_noise)
        img=self.remove_line(p1,p2,LL,img_removed_noise)
        img=median_filter(np.asarray(img),1)
        return img

    def remove_noise(self,img):
        img_gray=img.convert('L')
        w,h=img_gray.size
        max_color=np.asarray(img_gray).max()
        pix_access_img=img_gray.load()
        row_img=list(map(lambda x:255 if x in range(max_color-15,max_color+1) else 0,np.asarray(img_gray.getdata())))
        img=np.reshape(row_img,[h,w])
        return img

    def apply_median_filter(self,img):
        img_gray=img.convert('L')
        img_gray=cv2.medianBlur(np.asarray(img_gray),3)
        img_bw=(img_gray>np.mean(img_gray))*255
        return img_bw

    def eliminate_zeros(self,vector):
        return [(dex,v) for (dex,v) in enumerate(vector) if v!=0 ]

    def get_line_position(self,img):
        sumx=img.sum(axis=0)
        list_without_zeros=self.eliminate_zeros(sumx)
        min1,min2=heapq.nsmallest(2,list_without_zeros,key=itemgetter(1))
        l=[dex for [dex,val] in enumerate(sumx) if val==min1[1] or val==min2[1]]
        mindex=[l[0],l[len(l)-1]]
        cols=img[:,mindex[:]]
        col1=cols[:,0]
        col2=cols[:,1]
        col1_without_0=self.eliminate_zeros(col1)
        col2_without_0=self.eliminate_zeros(col2)
        line_length=len(col1_without_0)
        dex1=col1_without_0[round(len(col1_without_0)/2)][0]
        dex2=col2_without_0[round(len(col2_without_0)/2)][0]
        p1=[dex1,mindex[0]]
        p2=[dex2,mindex[1]]
        return p1,p2,line_length

    def remove_line(self,p1,p2,LL,img):
        m=(p2[0]-p1[0])/(p2[1]-p1[1]) if p2[1]!=p1[1] else np.inf
        w,h=len(img),len(img[0])
        x=list(range(h))
        y=list(map(lambda z : int(np.round(p1[0]+m*(z-p1[1]))),x))
        img_removed_line=list(img)
        for dex in range(h):
            i,j=y[dex],x[dex]
            i=int(i)
            j=int(j)
            rlist=[]
            while i>=0 and i<len(img_removed_line)-1:
                f1=i
                if img_removed_line[i][j]==0 and img_removed_line[i-1][j]==0:
                    break
                rlist.append(i)
                i=i-1
            i,j=y[dex],x[dex]
            i=int(i)
            j=int(j)
            while i>=0 and i<len(img_removed_line)-1:
                f2=i
                if img_removed_line[i][j]==0 and img_removed_line[i+1][j]==0:
                    break
                rlist.append(i)
                i=i+1
            if np.abs(f2-f1) in [LL+1,LL,LL-1]:
                rlist=list(set(rlist))
                for k in rlist:
                    img_removed_line[k][j]=0

        return img_removed_line

if __name__ == '__main__':
    image = cv2.imread("captcha.png")
    img = Image.fromarray(image)
    p = preprocessing()
    imgNew = p.pre_proc_image(img)
    cv2.imshow("Input", np.array(image))
    cv2.imshow('Output', np.array(imgNew, dtype=np.uint8))
    cv2.waitKey(0)

The code has no errors however the output image has none of the lines removed and looks somewhat like this:

I want the output to be completely free of any form of lines or at least reduce their intensity so that later it can be used with pytesseract to identify the letters mentioned in the captcha.

Update

There were few anomalies in the captcha data set where the lines had the same intensity such as given below

And after thresholding these images yet had some lines in them

After surfing the net I found that you can use techniques of erosion and dilation on these images to remove such lines however using these techniques, pytesseract is not able to recognize these characters since I do not get a very clear output.

Are there any other suggested techniques which can be applied for these sets of images so that later pytesseract can identify these characters ?

Answer 1

In this special case it seems density of lines is less than characters density. so by applying some thresholding methods you can remove line:

For example the following line give you this:

retval, image = cv2.threshold(image, 12, 255, cv2.THRESH_BINARY)

later by applying some noise removal methods, like median (from your own code), you can get this result: