Character Recognition using deep learning OpenCV python

Published in

Eazy Ciphers

4 min readAug 14, 2020

An eazy code from eazy ciphers

Hii!! welcome you all to our first medium story…. Well when I work on image datasets, I always think why these characters displayed in the image can be retrieved in a simple text format.

So, with that inspiration in mind to retrieve the data present in images. I surfed on the internet, I found a lot of ways to find and extract or recognize the characters present in an image.

In that surfing for character recognition, I found some online tools which will take input from the user and give the output.

The same way, we thought to simply in much better eazy way that any beginner can understand it.

Character Recognition:

Character Recognition process helps in the recognition of each text element with the accuracy of the characters.

The accuracy of classification and identification of characters or text elements in an image is measured by applying deep learning algorithms. These characters make a lot of difference in literature.

When we apply normal random techniques for recognition of characters may give some errors at a particular point. If we use deep learning OpenCV algorithm will give an effective output.

For the first way of approach to run our model, one has to install tesseract which is developed by Google as an optical character recognition engine.

pip install pytesseract

OpenCV:

Open CV is a deep learning technique where we mainly used to execute the data by calculating many eliminations of noises.

Open CV is the most popularly used algorithm in the branch of deep learning techniques.

It makes the colossally reliant on the data appeared from the trained characters and helps in recognizing the text present in the image. Open CV makes the missing accuracy of an algorithm into a picture. Gray scaling gives a valid classification in the field of character recognition. So hereby we import all the necessary packages for making our model into a good track.

In order to install OpenCV in your local computer use this command...

pip install opencv-python

Can an image be directly read from the code?

Let's discuss what is affecting to recognize the characters:

Noising in an image causes many factors of false recognition of characters. for ensuring noise-free in our recognition eliminate in code.
When the image is not high resolution then identification fails. So, it is better to take a high-resolution image for accuracy in the result.
Sometimes the angle of the image also defects.
Reflection of an image makes incorrect in the assumption of text. It makes to identify the letter “F” as “P” like that so many.
Fonts also vary the results if the code isn’t able to train it.
Various images have various styles of representation of the art, so, when there is more color complexion or multiple colors make incorrect assumptions of recognition text in an image.

Considering all the above factors in mind one have to build the code accordingly where all the test cases passes.

Python code for recognizing characters using OpenCV:

This code can be downloaded for your easy understanding of approach to the recognition.

Importing all the packages:

#import all the packages
import cv2
import numpy as np
import pytesseract
from PIL import Image

once you install with the packages import those into code.

Declare or initialize the path:

A tesseract is an open-source tool it can be downloaded online. After downloading mention its path like below.

# path of pytesseract execution folder 
pytesseract.pytesseract.tesseract_cmd = 'C:\\Program Files\\Tesseract-OCR\\tesseract.exe'# Path of image
main_path = r'qu12.png'

Main Method:

In this part of code, we are implementing

Read the image
Greyscale the image
Do dilation and erosion to eliminate unwanted noises in the picture.
Write the image after the removal of the noise.
Apply the threshold to get the only black and white pictures.
Write the same image for the further recognition process.
Character recognition with Tesseract.

def get_string(pic_path):
    # Reading picture with opencv
    pic = cv2.imread(pic_path)# grey-scale the picture
    pic = cv2.cvtColor(pic, cv2.COLOR_BGR2GRAY)# Do dilation and erosion to eliminate unwanted noises
    kernel = np.ones((1, 1), np.uint8)
    pic = cv2.dilate(pic, kernel, iterations=20)
    pic = cv2.erode(pic, kernel, iterations=20)# Write image after removed noise
    cv2.imwrite(main_path + "no_noise.png", pic)#  threshold applying to get only black and white picture 
    pic = cv2.adaptiveThreshold(pic, 300, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 31, 2)# Write the image for later recognition process 
    cv2.imwrite(main_path + "threshold.png", pic)# Character recognition with tesseract
    final = pytesseract.image_to_string(Image.open(main_path + "threshold.png"))return final