(Archived) Request: Correcting text parsed from images?

[soundsgoodtome 155]

What evernote does is truly amazing, and I'm grateful to have this tool in my arsenal. However, sometimes the text-recognition is (understandably) inaccurate.

Today I was searching for a photo I took of a bag of mulch that was used in my yard. (Oh, the uses for evernote!) I searched for "mulch" and found the image, but evernote also brought up two other notes that had nothing to do with "mulch." Turns out both images came from a web site and contained images with the word "music," written in a fancy, script font that confused evernote.

Again -- totally understandable, but it occurred to me that it would be nice to be able to correct evernote and tell it that those words are "music" not "mulch." This would be particularly useful for future searches when, in fact, I'm looking for "music."

[jbenson2 2,147]

Today I was searching for a photo I took of a bag of mulch that was used in my yard. (Oh, the uses for evernote!) I searched for "mulch" and found the image, but evernote also brought up two other notes that had nothing to do with "mulch." Turns out both images came from a web site and contained images with the word "music," written in a fancy, script font that confused evernote.

Again -- totally understandable, but it occurred to me that it would be nice to be able to correct evernote and tell it that those words are "music" not "mulch." This would be particularly useful for future searches when, in fact, I'm looking for "music."

Here is a two step work-around that I use.

I'll use a complicated search to show how it works.

Search a notebook named HOME for the word MULCH in notes created during the PAST COUPLE MONTHS

• Notebook:Home Mulch created:month-1

Step 1.) When I find a wrong match, I add a tag X to the note.
Step 2.) I modify the search by removing the X tag notes with the -tag comment

• Notebook:Home Mulch created:month-1 -tag:x

[BurgersNFries 2,407]

Again -- totally understandable, but it occurred to me that it would be nice to be able to correct evernote and tell it that those words are "music" not "mulch." This would be particularly useful for future searches when, in fact, I'm looking for "music."

I often add "keywords" to notes to facilitate the search. In the above case, you'd just go into that note & type in the word "music" so it will show up when you do a search on "music."

[soundsgoodtome 155]

I often add "keywords" to notes to facilitate the search. In the above case, you'd just go into that note & type in the word "music" so it will show up when you do a search on "music."

Yes, I've done that, too. The only caveat is that the music files still come up, when searching for "mulch." But this is a good stop-gap measure.

Thanks.

[BurgersNFries 2,407]

Yes, I've done that, too. The only caveat is that the music files still come up, when searching for "mulch."

Jbenson suggested his method of dealing with that. I'd guess that's a much more reasonable solution than EN allowing you to change their OCR results (which sounds like a big deal thing to me & so not something they would consider doing in the near future, if ever.)

[jbenson2 2,147]

My original suggestion was complicated on purpose to show the capabilities of search.

Here is a simplified search for the word MULCH across all notebooks

• Mulch

Step 1.) Any mis-matches? add a tag X to those notes.
Step 2.) Run search again with the -tag comment (to hide the mis-matches)

• Mulch -tag:x

[BurgersNFries 2,407]

If I'm understanding correctly, the one problem you could have with this method is if you had a third image that had the word "mulch" in it but EN OCRd it as "myth". You search on "myth" & when the third "music" image shows up, you tag the third image with the X tag and type the word "mulch" into the note. The next time you do "Mulch -tag:x", the third image won't show up. But there's a certain point where you just have to visually filter through the results anyway, since most searches will yield multiple results, if you have a very large database and you're doing a fairly simplistic search. IE, if you're doing a search only on the word "music" and your database is very large, there may be several notes that the search yields & you'll have to manually review the titles or notes to find the one(s) you are looking for. The way around this is to refine your notes & searches with notebooks, mutiple tags & titles. But even then, (at least IME), you can end up with a results list of multiple notes that you have to manually filter through. That's where descriptive/accurate titles help, too. IE, "favorite mulch" for the mulch bag. "mulch - not myth" for the third image I mentioned above in this post. This sounds much more complex than it really is. I make it a point to make sure all notes have an accurate & descriptive title & add in any additional "keywords" I think I may use in the future. IE, if a note pertains to our fixed assets software (by FAS Sage), I make sure the words "fixed assets FAS sage" are all in the note. Then if I search for FAS or Sage or "fixed assets", that note will appear. Those two things (titles & keywords) help a lot, when doing searches. Notebookss & tags help you further refine your searches.

Having said all that, I guess my point is that if you use the tools already in EN, it won't matter that the "music" image showed up. If you title the image of the mulch bag "favorite mulch - buy this kind" or "hate this mulch - AVOID" and if you have many other notes with the word "mulch" (IE 20090901 - added mulch to the yard", "need to buy more mulch" and maybe even one where the word "much" is spelled incorrectly as "mulch", then you can quickly look at the titles & find the one you're looking for.

[jbenson2 2,147]

Agree 100% with the importance of well written titles. They can be incredibly powerful and useful in searches.

I'll use the web headline when I do a webclipping.

But for any note that I consider important and/or filed in my Personal notebook, I try to use a standard format.

• Date / Location / Subject / Individuals / Timecode

• 20090911 USA MN Rogers Dinner at Maynards receipt JLB DLB 145435.pdf

Date: 20090911

Location: USA MN Rogers

Subject: Dinner at Maynards receipt

Individuals: JLB DLB

Timecode: 145435 (from my scanner)

To find all my notes for Minnesota: intitle:"USA MN"

To tighten up the search for all my notes for Rogers, Minnesota: intitle:"USA MN Rogers" or intitle:"MN Rogers"

I try to use standard abbreviations whenever possible. Mpls for Minneapolis