You snap images of the reports using your smartphone, and a previously daunting task is being accomplished effortlessly. In no time, what used to be big piles of paper reports are now digital data on your computer. So, an initially overwhelming task turned out to be smoothly handled, allowing you to digitize heaps of old reports swiftly and efficiently.

Sound like some fictional movie ? right! Well it's not . welcome to world of Optical Character Recognition, commonly known as OCR. 

Optical Character Recognition (OCR) is the process of detecting and reading text in images through computer vision. OCR software processes the characters in such a way that a computer can now read and recognize text: letters, symbols, words, etc.

It is the recognition of text from printed or handwritten documents and images in order to distinguish alphanumeric characters using technology. 

Image Source: Buff ML

Now, here's the thing. When we humans look at something, our brains instantly grasp it. We see dark and light regions, notice shapes, and in a millisecond, our minds process it all. Computers, on the other hand, see everything as just a bunch of 0's and 1's, lacking the patterns, what our human eyes catch effortlessly. This is where OCR steps in, translating those patterns of 0's and 1's into machine-readable data.

How Does OCR Work? 

Firstly, imagine picking up a cookbook from a chaotic bookshelf filled with haphazardly placed books of every size and shape. Sounds complicated, right? That's similar to how OCR feels when it first encounters a document. It needs to figure out the structure. So, whether it's a PDF with text in lines or split into two parts, OCR adjusts accordingly. OCR needs to adjust its "reading" glasses accordingly. It's like making sure OCR knows how to read receipts from a grocery store or slips of banks. 

Alright, now we've picked the recipe, but hey, the cookbook is dusty. We need to get the pages ready for reading - here's where 'pre-processing' comes in. We brush away all the dust (or 'noise') from the image, sharpening the text. In this stage, We work on cleaning up the document by getting rid of any unnecessary stuff in the image like noise in the image. In the stage, OCR can clearly see which parts are the 'characters' (black text) it needs to recognize, and which parts are just the background (white paper).

Now we have a clean recipe book, let's start cooking, or in OCR terms, let's begin recognizing text! This process can happen in two ways, pattern recognition or feature detection.

Think of pattern recognition like your brain matching photos of your favorite dish to the corresponding recipe. In the same way, the OCR looks the similarity between the character present in image, with the set of known fonts and symbols. This method works great for straightforward text, but it can run into issues if there are unique characters or symbols that aren't part of the standard set.

The second method involves recognizing text through feature detection algorithms. Here, OCR tries to understand the text in a more creative way. It looks at things like shapes, the number of lines, or even how many triangles are present. This approach works well for unusual fonts or characters. However, for regular text, the standard pattern recognition method is usually the right way to go.

Image Source: Anastasiia Molodoria on Mobidev

Fantastic! We've cooked up a digital document, but let's taste-test before serving. If something tastes off, we adjust it, right? In the same way here, Any difference between the image data and extracted data is adjusted either manually or via script.

This difference can occur because of two reasons, either because image is too noisy and difficult to understand. The second reason can be misunderstanding to the OCR because of two or more similar type of symbols. For example, OCR many times confuse between “1” and “|” or “O” and “0” or “S” and “5”.

Types of OCR.  

There's a whole array of OCR tools you can find out there, all thanks to the advancements in technology. Let's have a look at some popular ones - Adobe Acrobat Pro DC, ABBYY FineReader, Amazon Textract, and Tesseract.

But hey, in this blog post, we won't be diving into these. Instead, we will focus on some awesome open-source OCR tools that won't cost you any money! Exciting, right?

EasyOCR 

EasyOCR is an open-source OCR tool library implemented in Python. It can recognize more than 70 different languages including English, Hindi, Chinese, Korean etc. It can read text irrespective of the text orientation in the image.

 It is a very simple OCR, from a few lines of code, we can recognize text. Unlike many OCRs that read text line by line, EasyOCR can detect long lines of text. It supports the use of the GPU for fast processing, which becomes extremely beneficial for processing large volumes of images.

1. English text OCR

Image Source: Shutterstock

Text After OCR : The future belongs to those who believe in the beauty of their dreams.

-Eleanor Roosevelt

2. Hindi Text OCR 

Image Source: hindionlinejankari.com

Text After OCR : अगर तुम सूरज की तरह चमकना चाहते हो तो पहले सूरज की तरह जलो|

डॉ. एपीजे अब्दुल कलाम  

PaddleOCR 

Built on the strong foundation of PaddlePaddle, PaddleOCR is another impressive tool in the OCR world. It stands out with its extensive language support - over 200 languages Plus, it offers a library of over 80 pre-trained models ready to tackle tasks around text detection, recognition, and layout analysis. PaddleOCR truly shines is in handling complex text detection tasks, like table layouts. It's a top contender among all the open source OCR for all your complex text detection needs.

The way PaddleOCR operates is quite straightforward - a three-step process. Firstly, it detects potential text within a document and encapsulates it within a bounding box. Secondly, if the text isn't horizontally aligned, PaddleOCR takes care of rotating it to the standard orientation. Finally, it identifies each character within the bounding box, resulting in accurate text recognition.

Examples 

1. English text OCR 

Image Source: Pinterest

Text after OCR : IF YOU CAN STAY POSITIVE IN NEGATIVE SITUATION YOU WIN.

2. Hindi Text OCR

Image Source: Carrerindia

 Text after OCR : जिस दिन आपके सामने कोई समस्या न आये आप समझ लें कि आप गलत रास्ते पर जा रहे हैं|

Tesseract

It is an OCR engine developed by Google. It is considered as one of the most accurate open-source OCR engines available.

Image Source: Filip Zelic & Anuj Sable on Nanonets

It supports more than 100 languages. It has the specialty that it can be trained to recognize other languages. It also allows us to train it on new fonts. This feature sets it apart from many other free OCR engines. It also performs page layout analysis which helps in maintaining the structure of the document during the OCR process. It has strong community support that continually contributes to its improvement and development.

Examples 

 1. English text OCR 

Image Source: iStock

Text after OCR: FAILURE is not the opposite of success it‘s PART OF SUCCESS

2. Hindi Text OCR 

Image Source: bodhvichar

Text after OCR : असफल होने पर आप निराश हो सकते हैं, लेकिन यदि आप कोशिश नहीं करते हैं तो आप बर्बाद हो जाते हैं।

- बैवर्ली सिल्‍स

Conclusion

OCR is a powerful tool, acting as a bridge between physical documents and digital data. We looked at three prominent open-source OCR tools: EasyOCR, PaddleOCR, and Tesseract, each with its unique strengths and areas of focus. EasyOCR shines with its simplicity and support for long lines of text. PaddleOCR impresses with its feature of handling complex texts. Lastly, Tesseract stands out with its exceptional accuracy, speed, and ability to be trained for new languages and fonts.

Choosing the best OCR tool depends on specific needs. EasyOCR could be a preferred choice for beginners, while PaddleOCR could be use for complex text and fonts. Tesseract could be an excellent choice for large-scale processing or for the OCR of standard text.

dreamstime.com, Filip Zelic & Anuj Sable on Nanonets, Carrerindia, Pinterest, hindionlinejankari.com, Shutterstock