How to extract only the date from a food label image?

I am looking to create a program that reads only the date from a product label such as:

I have currently tried using OCR and bounding boxes to see which part I'm isolating but I have had no luck just isolating the date which can be used on multiple images.

Any idea's about how best to go about this would be helpful.

Many thanks.

Answer 1

[Cris LaPierre]

Is the label an image? I'd suggest looking at this example on optical character recognition.

Code up what you can, and then come back with specific questions about any parts you are getting stuck on.

[Stevie]

Yes the label is an image, sorry.

I have tried the OCR text methods, it reads all of the text, i'm trying to isolate only the expiry date, I wasn't sure what method to use to do this. Thanks.

[Cris LaPierre]

Try this:

% Load an image 
I = imread('productLabel.jpg');

% Perform OCR
results = ocr(I,'CharacterSet','0123456789ABCDEFGJLMNOPRSTUVY');

% Display the date
date = [results.Words{3} ' ' results.Words{4}]

date = '10 JUN'

The 0 was getting recognized as [] so I set the character set to be all numbers plus all letters found in the 3-letter month abbreviations.

Is there a particular datatype you want to date returned, or is a string ok?

[Stevie]

This is a good idea, thank you for that.

The string could be converted to a date datatype, I should be able to find that information. What would you recommend to automatically extract the date, rather than having to specify the "results.words{}" ?

[Cris LaPierre]

date = datetime(datevec(date))

date = 
   10-Jun-2018 00:00:00

Note that this will make assumptions about the year. You will likely need to develop some code to control this if year is important.

What is the variablility in the labels you are looking at? Are they all like the image you uploaded?

[Stevie]

I'm looking at using a webcam to take images of food products (main image position focus will be on the expiry date) but sometimes other info is included. Like the image below:

[Cris LaPierre]

Again, it's going to depend on the format of your labels. Based on the 2 you've shown, here's what I'm thinking.

1. Use 'ismember' to determine which word is the month
2. Use the bounding box y-position to find the other words on the same line
3. Put the words together to recreate the date
4. Convert the string to a datetime

This works on the 2 images you've uploaded

% Create list of months
months = {'JAN','FEB','MAR','APR','MAY','JUN','JUL','AUG','SEP','OCT','NOV','DEC'};
% Find word that is a month (only expecting 1)
idx = ismember(word,months);
% Extract Y position of month
y = results.WordBoundingBoxes(idx,2);
% Find other words that are on the same line (+/- 0.1*height of month
sameY = abs(results.WordBoundingBoxes(:,2)-y)<results.WordBoundingBoxes(idx,2)/10;

% Convert to datetime
date = horzcat(word{sameY})
date = datetime(datevec(date))

[Stevie]

Many thanks for all of your help! you've been a great help.