OCR Nutrition Information, Ingredients & Quantity from Food Packet Images

  • 89 Views
  • Last Post yesterday
Austin Christopher posted this 2 weeks ago

I am trying to implement OCR solution provided by ABBY as I found better text recognition in images when I tried testing in the url : https://cloud.ocrsdk.com/Demo/

But when I try to test the same locally using Abby.Ocrsdk.client (Java) implementation. Not able to get reliable results.

Please let me know steps to test locally for best results.
Also I would want to know how the response structure looks like.

Awaiting your response.

Also wanted to know if there is any feature of translation of content to English provided by ABBY?

Should Images be preprocessed before OCR? 

Order By: Standard | Newest | Votes
Helen Osetrova posted this 4 days ago

Hello Austin,

 

1) As for your first question, what kind of issue do you face? What form does it take? 

Please review our Quick Start Guide to learn how to form a request to ABBYY server and how to get a result of processing. You could also find helpful the following article. It describes not only using Cloud OCR SDK under Linux but the main stages of configuring the Java sample as well. 

 

2) Cloud OCR SDK does not provide any tools for content translation, its main purpose is recognition. For such option please try our ABBYY Lingvo Live service. Please note that ABBYY Lingvo Live is not an SDK but ready-to-use dictionary, so it is not possible to automate translation process with its help.

 

3) No special preprocessing is needed if source images comply our Source Image Recommendations. Cloud OCR SDK is able to correct orientation, skew and some of distortions of source image automatically. If your images are of good quality, you can even skip some of the preprocessing stages to increase recognition speed. 

 

Hope this information will be helpful!

 

Austin Christopher posted this 2 days ago

Hi Helen,

 

Thank you for your quick response.

My request here is, I wanted to know if the API can be customized to identify nutritional information listed below from food packet images:

 

  1. Net weight (g or ml);
  2. Energy value per 100 g (or per 100 ml);
  3. Content per 100 g (or per 100 ml) of each of the nutrients in the label (fat, saturates, carbohydrate, total sugars, protein and salt); 
  4. Content per 100 g (or per 100 ml) of the items in the voluntary declaration (fibre, polyols and others)*;
  5. Ingredients (individualized in a sub-table, order number of presentation, its percentage in the product)*;
  6. Portion size;
  7. Package size36*;
  8. Nutrition and health claims*;

 

 

 

Helen Osetrova posted this yesterday

Hi Austin,

 

You should be able to obtain all the required information as a text and then classify the fields on your side.

To do this, please try to specify the imageSource=photo and profile=textExtraction of the processImage method. Please note that the photos of the food packets should have a good readable quality, without glares and text distortions. Please review the corresponding article to know more about source image requirements. 

 

Hope it will help!

 

Close