Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I'm scanning in many A4 paper documents to JPEG using a automated document feeder scanner.

The results are FILE0001.JPG FILE0002.JPG etc.

I would like a program to rename the file name title to text found in the actual scanned jpeg image itself. Preferrably to determine the title, the program would look for the largest text in the image and which appears closest to the top of the image.

I am aware of several commercial and some free OCR applications and would be willing to purchase if necessary, however these appear to have more than what I need: they convert to PDF etc. whereas I would just prefer to keep it simple and work with the original scanned image.

Would welcome out-of-the box easy to use programs for Windows XP, 7 or MacOS.

share|improve this question
add comment

1 Answer

What you are looking for is something that is extremely complex and unlikely to be reliable even if it could be found I'm afraid.

I think that the best you could hope for would be to make use of either Microsoft OneNote (part of Office) or Evernote (has a free version).

These are both able to OCR images in notes - in the background - leaving any discovered text searchable. I'm not sure whether they would pick up the note title from the text though, they might if you make sure that no other text is in the note. Give them a go.

Be warned though that OCR even of well-scanned typed or typeset documents is far from reliable and even then, knowing what constitutes a title, though easy for humans to parse, is a very hard task for a computer.

UPDATE: The complexity comes from a number of things. The act of OCR'ing an image to text is complex enough for a machine to do. There are so many complexities to language that it is very difficult to pick out meaning from an image even when that image is typeset. Even typeset characters vary massively, especially when scanned due to scanning limitations, changes of angle, smudged or otherwise damaged source text (e.g. a fold in the paper) and so on. Secondly, what is a title? Obvious, you might think - something of a larger size than "average" towards the top of the page? How does the system work out the average font size? Itself a significant task as it needs to "parse" the whole scan. Then there are many combinations of layout - which ones should the machine try to recognise? Take an average business report for example, it may have several title-like text elements.

Each of these bits of processing are going to take significant time on even a modern PC and involve large amounts of data processing: Clean the image, straighten the image (recognising edges and "lines" of text), pick out font styles to understand what is text and what isn't, attempt to recognise the text (probably applying spelling and grammar rules), work out the font sizes and average, identify repeating elements (headers/footers) to ignore, try to identify larger text early in the document. Guess the title, check if it is a valid file name for the platform, change if not, ensure name is unique and unused. Phew!

At best, most OCR tools aim for around 90% accuracy from standard scans with clean, straight-fed documents. Do you want to have 90% accurately titled documents? That might be OK to you but would customers of a product put up with it? After all of the development, would the risk be worth it to vendors?

I don't know the answers to these questions. I can see that it could be a great feature but I'm not aware that anyone offers this (I've done a quick check via Google too).

It would be easier if all of your documents are the same layout. Then you could use "zoning", something that most of the better tools offer and take the appropriate zone as the basis for the file name. This would be more (but not completely) reliable. Perhaps you should check with some of the vendors to see if they are interested in doing this.

share|improve this answer
    
+1 Thanks Julian Knight, for the Evernote and for the comments on reliability. I'll leave it open for other to contribute. –  therobyouknow Jan 19 '13 at 13:28
1  
No problems. I hope you do get a better answer, I'll monitor this thread to see if you do as I'd be interested too. –  Julian Knight Jan 20 '13 at 11:48
    
+1 Thanks Julian Knight. I hope to close and accept an answer (perhaps yours!). I'd welcome further thoughts on why this might be complex as, if the OCR can recognise the text, it should know about the 'bounding box' around the text, i.e. it should be able to work out how much area the text is taking up and therefore the size. It should also be able to work out how close it is to the top of the page from its position. So yes, I would agree that OCR itself is complex & may not be 100% right all the time with 'busy' pages but once this bit is done,the co-ordinates and size should be easier,no? –  therobyouknow Jan 21 '13 at 13:34
    
This is really something for a discussion forum rather than here. But I'll try and add a little more info. –  Julian Knight Jan 21 '13 at 14:16
    
I probably should have put in a comment about some of the RECEIPT processors now available online. They offer auto-detection of vendor, currency and amount and are fairly amazing at their accuracy thanks to having plenty of back-end processing power available. –  Julian Knight Jan 21 '13 at 14:35
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.