System for extracting text from images

1. A system for extracting text from images, the system comprising at least one processor configured to:

receive a digital copy of an image;

identify a portion of the image using a first custom deep neural network that is trained by the system, wherein the portion comprises text to be extracted;

determine orientation of the portion of the image with respect to a horizontal axis using a second custom deep neural network that is trained by the system;

reorient the portion of the image horizontally based on the determined orientation of the portion of the image; and

extract text from the portion of the image considering the orientation of the portion of the image, wherein the processor is further configured to:

extract images of relevant characters from the portion of the image comprising the relevant characters that are superimposed over irrelevant characters, with each extracted image of relevant characters comprising individual relevant characters, wherein at least a set of relevant characters is at a different orientation as compared to the irrelevant characters.

2. The system as claimed in claim 1 , wherein the portion comprises a second image, which is distinguishable within the image, wherein the second image comprises the text to be extracted.

3. The system as claimed in claim 2 , wherein the first custom deep neural network is trained by configuring the system to:

receive a plurality of training images with pre-identified portion of interest within each of the training images;

predict portion of interest in each of the training images; and

refine the first custom deep neural network based on the pre-identified portion of interest and the predicted portion of interest corresponding to each of the training images.

4. The system as claimed in claim 2 , wherein the second custom deep neural network is trained by configuring the system to:

receive a plurality of orientation training images with pre-identified orientation corresponding to each of the orientation training images;

generate a set of training images using the received plurality of orientation training images;

predict orientation of each of the orientation training images; and

refine the second custom deep neural network based on the pre-identified orientation and the predicted orientation corresponding to each of the orientation training images.

5. The system as claimed in claim 1 , wherein extracting the text from the portion of the image is enabled by configuring the processor to:

classify character present in each of the extracted images of the relevant characters into character groups; and

determine sequence in which each of the characters is presented in the portion based on co-ordinates of the extracted images of the relevant characters within the portion.

6. The system as claimed in claim 5 , wherein the irrelevant characters comprises:

a first set of characters, which are oriented in the same direction as that of the characters in the extracted images of the relevant characters; and

a second set of characters, which are oriented in one or more directions that are different compared to the direction of orientation of the characters in the extracted images of the relevant characters.

7. The system as claimed in claim 5 , wherein extraction of images comprising individual characters is enabled by a third custom deep neural network, wherein the third custom deep neural network is trained by configuring the system to:

receive a plurality of labelled training images with pre-identified areas within each of the labelled training images, wherein each of the areas binds one character;

predict areas, each comprising one character, in each of the labelled training images; and

refine the third custom deep neural network based on the pre-identified areas and the predicted areas corresponding to each of the labelled training images.

8. The system as claimed in claim 5 , wherein classifying character present in each of the extracted images into character groups is enabled by a fourth custom deep neural network, wherein the fourth custom deep neural network is trained by configuring the system to:

receive a plurality of character images corresponding to each of the character groups with pre-identified character for each of the character images;

predict character presented in each of the character images; and

refine the fourth custom deep neural network based on the pre-identified character and the predicted character corresponding to each of the character images.

9. A method for extracting text from images, the method comprising:

receiving, by a computing infrastructure, a digital copy of an image;

identifying, by the computing infrastructure using a first custom deep neural network that is trained by the computing infrastructure, a portion of the image, wherein the portion comprises text to be extracted;

determining, by the computing infrastructure using a second custom deep neural network that is trained by the computing infrastructure, orientation of the portion of the image with respect to a horizontal axis;

reorienting, by the computing infrastructure, the portion of the image based on the determined orientation; and

extracting, by the computing infrastructure, text from the portion of the image considering the orientation of the portion of the image, wherein extracting text is performed by extracting images of relevant characters from the portion of the image comprising the relevant characters that are superimposed over irrelevant characters, with each extracted image of relevant characters comprising individual relevant characters, wherein at least a set of relevant characters is at a different orientation as compared to the irrelevant characters.