Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

OCR library to recognize numbers on meter device

I need OCR library(preferably in Java), which will be able to determine digit visualized on meter devices. Please see image. I've tried some libraries in Java but they was not able to determine this kind of font.

enter image description here

UPDATE 1: Tesseract and Asprise libraries failed to perform this task

like image 230
narek.gevorgyan Avatar asked Jun 20 '12 06:06

narek.gevorgyan


People also ask

How accurate is OCR meter reading?

Fast and Accurate OCR Meter Reading. Mobile meter scanning eliminates the errors of manual data entry, and reduces the time needed to get readings by 90%. Just point the device and start scanning! Anyline will precisely read multiple generations of meters and reliably document them in your system.

What is OCR and how does it work?

Time-saving and Increased Efficiency OCR helps decrease the amount of time it takes to read the meter and input the data accurately. As a result, the OCR boosts employee efficiency, meaning fewer hours and a cost-saving opportunity. In addition, the rate is around 20x faster than manually reading the meter and entering the data.

Can OpenCV be used to read electricity meters?

OpenCV practice: OCR for the electricity meter. OpenCV (Open Computer Vision) is a powerful and comfortable environment for the realization of a variety of projects in the field of image processing. This tutorial introduces some aspects of OpenCV based on a practical application - the reading of an electricity meter.

How can machine learning be used to read utility meters?

Our machine learning technology can automate the process of reading utility meters and performing data entry via our APIs and SDKs. An image speaks more than a thousand words. Below is an example of the three steps our OCR engine takes to automatically extract utility meter data from documents and photos.


1 Answers

You could take a look at Tesseract, it's a Google maintained open source OCR. However, I think that if you just supply the image as it is the OCR might have some issues, reason being that you have digits which vary in size, and this will most likely cause issues while the Neural Network(s) are learning.

So, I think that you would have a larger chance of success if you filter your image first and try to remove items that you do not need to identify and try to make the digits you need to read the same size.

EDIT: As per your edit, you can also take a look at Java OCR:

Java OCR is a suite of pure java libraries for image processing and character recognition. Small memory footprint and lack of external dependencies makes it suitable for android development. Provides modular structure for easier deployment

That being said, I still think that you should do some processing on your image first.

like image 56
npinti Avatar answered Sep 18 '22 12:09

npinti