OCR

 Introduction

Optical character recognition (OCR) is the electronic identification and digital encoding of typed or printed text by means of an optical scanner and specialized software.

About the Project

In today’s world there is a huge use of electronic document than physical documents. For that OCR is one of the best technologies for reading scanned image and convert it into a digital character-based format. 

We know that there are so many OCR available for the English language with the high accuracy, but for the Gujarati language there is OCR available but not that much accurate. It is very difficult to find proper Gujarati printed alphabets separately. So, for Gujarati OCR there is still work going on. 

These are the basic alphabet of the Gujarati language that is used in the sentences and to talk all around the Gujarat state. 

These are the vowels used with alphabet k. There is a total of 11 vowels. We are going to collect all the possible characters of different size, shape, fonts. After collecting characters, we can make one database of that and we can train a model then we can use it to identify the Gujarati alphabets.

Methodology

We have started with writing Gujarati alphabets using fonts like Hari, Shruti etc. of size 100. After writing all the alphabets with their vowels. We have taken the screenshots of the printed alphabets with their different variations. 

Here is the sample of the alphabets that we have taken by applying noise, blur, resize function to that image.

We can use Matlab or photoshop also to apply this kind of function to the image. Here we have used Matlab for applying a different kind of filters. We have to just give the path of the folder and then it will automatically take the image from that folder and after applying the method on it it will be stored in the folder that we have given in that code. It will generate around 300-400 different type of image for the single alphabet. 

Code of Matlab:

Here in these two codes of line we have read the image using imread () from the particular location and using imresize () function we can resize the image.

We have used imwrite () function to save that image in the same folder from where we have read the image

So, we have used different noise function to add the noise into the image and also used blur methods to blur the images. After that using resize function we have resized the image and again apply the all the function on the image and it will again generate images according to that.

After collecting this we are going to train the data that we have generated in the form of an image. There are around 1,36,000 images dataset.

We can train our model using different available code and all. We have to just give the input as image and it will train the model according to our need.

We have to divide our dataset into our two parts like train dataset and test dataset into a portion of 70% and 30% respectively. So, after training the dataset we can test it.

Conclusion

From this project we came to know about different functions of the MATLAB, using which we can generate images easily and directly store it into our selected folder. We also came to know about how to generate the dataset. And how we can use that in different work.

-by Meet Mistry and Aman Mundhva