Digit Recognition OCR in OpenCV-Python

Question

My question is about building a simple program that detect digits inside images, i did some researches and found this topic Simple OCR digits on stack and i found it very educational, so i wanted to us it for my own need.

My training data image is like:

The code i used to build the dataset is: (i did some modifications to Abid Rahman's code so it can hundle my case)

import sys

import numpy as np
import cv2

im = cv2.imread('data_set_trans.png')
im3 = im.copy()

gray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray,(5,5),0)
thresh = cv2.adaptiveThreshold(blur,255,1,1,11,2)

#################      Now finding Contours         ###################

contours,hierarchy = cv2.findContours(thresh,cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)

samples =  np.empty((0,100))
responses = []
keys = [i for i in range(48,58)]


for cnt in contours:
    if cv2.contourArea(cnt)>20:
        [x,y,w,h] = cv2.boundingRect(cnt)


        if  h>=10:
            cv2.rectangle(im,(x,y),(x+w,y+h),(0,0,255),2)
            roi = thresh[y:y+h,x:x+w]
            roismall = cv2.resize(roi,(10,10))
            cv2.imshow('norm',im)
            print "Begin wait"
            key = cv2.waitKey(1)
            key = raw_input('What is the number ?') #cv2.waitKey didnt work for me so i add this line


            if key == -1:  # (-1 to quit)
                sys.exit()
            else:
                responses.append(int(key))
                sample = roismall.reshape((1,100))
                samples = np.append(samples,sample,0)

responses = np.array(responses,np.float32)
responses = responses.reshape((responses.size,1))
print "training complete"

np.savetxt('generalsamples.data',samples)
np.savetxt('generalresponses.data',responses)

I used the same training data image as testing part, in order to get the best results accuracy and see if i am on the right way:

import cv2
import numpy as np
import collections

#######   training part    ############### 
samples = np.loadtxt('generalsamples.data',np.float32)
responses = np.loadtxt('generalresponses.data',np.float32)
responses = responses.reshape((responses.size,1))

model = cv2.KNearest()
model.train(samples,responses)

############################# testing part  #########################

im = cv2.imread('one_white_1.png')
out = np.zeros(im.shape,np.uint8)
gray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
thresh = cv2.adaptiveThreshold(gray,255,1,1,11,2)

contours,hierarchy = cv2.findContours(thresh,cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours:
    if cv2.contourArea(cnt)>20:
        [x,y,w,h] = cv2.boundingRect(cnt)
        if  h>=10:
            cv2.rectangle(im,(x,y),(x+w,y+h),(0,255,0),2)
            roi = thresh[y:y+h,x:x+w]
            roismall = cv2.resize(roi,(10,10))
            roismall = roismall.reshape((1,100))
            roismall = np.float32(roismall)
            retval, results, neigh_resp, dists = model.find_nearest(roismall, k = 1)
            string = str(int((results[0][0])))
            cv2.putText(out,string,(x,y+h),1,1,(0,255,0))


cv2.imshow('im',im)
cv2.imshow('out',out)
#cv2.waitKey(0)
raw_input('Tape to exit')

The result was as like that:

As you can see it's completely wrong.

I don't know what i'm missing or if it my case is more particular and can't be handled by this digit OCR system ????

If someone could help me by any idea

I notice that i am using python 2.7 open-cv 2.4.11 numpy 1.9 and mac os 10.10.4

Thanks

Answer 1

I found the right way, it needed just more customised code.

The same process before detecting countours :

gray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray,(5,5),0)
thresh = cv2.adaptiveThreshold(blur,255,1,1,11,2)

Not

gray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
thresh = cv2.adaptiveThreshold(gray,255,1,1,11,2)

And

cv2.rectangle(im,(x,y),(x+w,y+h),(0,0,255),2)

Not

cv2.rectangle(im,(x,y),(x+w,y+h),(0,255,0),2)

I get 99% accuracy, good beggining percentage

Thanks for you anyway