Number recognition

[jjlad]

Lightroom Version Number

10.1

Operating System

1. Windows 10

Since LR can detect and remember faces, would there be a way to get it to substitute numbers for faces and get LR to do number recognition? For sports shooters this would be a godsend! I'm almost thinking of taking some group photos and putting numbers where the heads are to see if LR would pick up the numbers as faces so I could label them as numbers. Has anyone tried anything relative to this??

 

[Hal P Anderson]

Lightroom won't do that. You could submit a feature request. See the link at the top of this page.

 

[clee01l]

Since LR can detect and remember faces, would there be a way to get it to substitute numbers for faces and get LR to do number recognition? For sports shooters this would be a godsend! I'm almost thinking of taking some group photos and putting numbers where the heads are to see if LR would pick up the numbers as faces so I could label them as numbers. Has anyone tried anything relative to this??

I don’t think you are going to have any luck. In Face Detection, Lightroom want to see “two eyes and a mouth” in the right relative location. It does not take much to mistake 3 birds in flight for a face.


Sent from my iPad using Tapatalk

 

[Paul_DS256]

I'd suggest that what you are looking for is general text recognition in images. Check out 'Any Vision'. There's an example there of reading a number off of a jersey.

Also see:

• Detect text in images | Cloud Vision API | Google Cloud
• Apple Developer Documentation

 

[Denis de Gannes]

I have my name identified on Facebook and other social media with other friends and relatives and the recognition is faulty. I live on a small island, I have this feature deactivated as it only causes slowdowns and confusion.

 

[Linwood Ferguson]

For sports shooters this would be a godsend! .... Has anyone tried anything relative to this??

I've tried using face recognition generally for sports shots, and for the most part it was a miserable failure. Angles, expressions, crowds, helmets. ... not sure exactly what, but the missed-faces and wrong-guesses were worse than doing it by hand.

What I've hoped for, and actually asked for without any reception, is context. I'd love to be able to tell facial recognition "here are the available people" (like by a keyword, which could be a team) and have it work with that limited set of faces. It seems like if it knew it was amount 30, instead of 3000 faces, it could do better.

Something like volleyball is not too bad. Soccer was awful. Baseball was maybe worse.

It's too bad, as number recognition would not work well for most of my shots, or at least a lot of my shots, as the numbers are not always on both sides, or otherwise hidden. Context-drive facial recognition though I think has real promise.

 

[johnrellis]

Check out 'Any Vision'.

Some of my customers have reported good results with Any Vision's recognition of bib numbers, others not so good. Commercial race photographers position their shots to get a clear, consistent view of the competitors and their bibs (and such bibs are designed for clarity in photos). But "action" photography (say of kids team sports) is more hit-or-miss, depending on how the jerseys are positioned in the shots.

 

[PhilBurton]

Since LR can detect and remember faces, would there be a way to get it to substitute numbers for faces and get LR to do number recognition? For sports shooters this would be a godsend! I'm almost thinking of taking some group photos and putting numbers where the heads are to see if LR would pick up the numbers as faces so I could label them as numbers. Has anyone tried anything relative to this??

I think that Photo Mechanic has that capability.

 

[Linwood Ferguson]

I think that Photo Mechanic has that capability.

To my knowledge all it can do is substitution in metadata, e.g. you enter #01 and it changes it to "John Smith, First Baseman for the Florida Frogs" or whatever. There are subscription services which provide the replacement values as well. I'm unaware of any image recognition there, but would be very interested if so.

 

[jjlad]

Thanks all. I have shot for Marathon Photo and they have a proprietary software that picks up numbers from hats and bibs and also records the colors of each runner's upper and lower clothing and hat. Then a few days after a marathon people can search the thousands of photos by any of those criteria ...or just the participant's name. Awesome. For some sports where I shoot the same teams repeatedly I create number name keywords and then just keyword all the photos to be able to find and sell them afterward. Not sure the work is worth the reward. Lucky when breaking even. I'll check out Any Vision.

 

[jjlad]

Just an update on this.
I did download and try Any Vision. I have backups of all the marathons I've shot so I tried it on numerous photos where the bibs had 3, 4 and 5 digit numbers. Didn't work at all. Didn't pick up a single number even though I had tweaked all the options to optimize it concentrating on text. Even text on banners and signs nearby was missed. Worse, if I let it add data automatically, I end up with lots I don't want.
Oh well ...it was worth a try.

I have the free Cloud Tagger installed and have used it on as many as one year's worth of photos all at once ...just starting it and walking away while it runs. It does not do text/number recognition but does OK on keywording if the confidence percent is set really really high ...like 97 or 98 percent. Otherwise hundreds of unneeded keywords get added overall (like scientific names for plants etc.). Since it works off the Google Photos program within Google, I can only imagine the keywording that goes on behind the scenes in that universe of data.

 

[johnrellis]

I did download and try Any Vision....I tried it on numerous photos where the bibs had 3, 4 and 5 digit numbers. Didn't work at all. Didn't pick up a single number

Hmm, could you send me a couple typical photos? (I'll private-message you.)

I hadn't tested race bibs for quite a while, so I grabbed some pseudo-randomly from Google Images. Any Vision (Google Cloud Vision) recognized all the unobscured numbers with no errors and only a couple false positives. This was quite a bit better than several years ago.

 

[johnrellis]

Cloud Tagger ... does OK on keywording if the confidence percent is set really really high ...like 97 or 98 percent. Otherwise hundreds of unneeded keywords get added overall (like scientific names for plants etc.).

Cloud Tagger and Any Vision both use Google Cloud Vision, so they should return the same keywords at a given score threshold. Note that Cloud Tagger is no longer available for download.

 

[jjlad]

Cloud Tagger and Any Vision both use Google Cloud Vision, so they should return the same keywords at a given score threshold. Note that Cloud Tagger is no longer available for download.

Interesting that Cloud Tagger isn't available any longer. Hope it doesn't have anything to do with the Pandemic.
Sorry John, I already uninstalled Any Vision because it didn't seem to work for me but I'm attaching 2 couple of examples of images on which I haven't removed all the tags Any Vision added. I think all those preceeded by a "<" were added by Any Vision.
Thanks for trying to help on this. Not sure why your copy of Any Vision would work and mine not work. Mac vs Windows 10?

 

[johnrellis]

I cropped your screenshots to low-resolution images (621 x 491 and 378 x 566) and Any Vision on Windows did a good job with them, correctly recognizing the clearly visible bibs:

Any Vision also extracted 1100 and 7355 from the first photo, which I think was Cloud Vision's attempt at the small blurry bibs in the background. With a full-resolution photo, it likely would have done better on those.

I configured the Advanced OCR option to extract just numbers with at least two digits:

If you also configured it to copy the extracted text to Caption or Title, you could search for bib numbers using the filter Text: Title Contains Words 11076.

 

[johnrellis]

Interesting that Cloud Tagger isn't available any longer. Hope it doesn't have anything to do with the Pandemic.

I think it may have been withdrawn prior to COVID. The author originally designed it for the careful-keywording use case, e.g. a stock photographer submitting relatively small numbers of photos to an agency, whereas Any Vision is designed for searching large catalogs. Cloud Tagger's pricing model reflected the use-case: The flat price for the plugin was intended to cover "lifetime" use of Google. This made sense for a typical user who keyworded 100 photos/month ($0.15) and was a lot easier on users compared to Any Vision (where you have to painfully sign up for a Google Cloud account). But that pricing wouldn't work with users who want to analyze 100K photos (the searching use-case). I believe Cloud Tagger later added the capability to analyze large batches of photos, but I don't think the author adjusted the pricing model accordingly.

 

[jjlad]

I must have missed something completely. I was looking in the keywords because after running Any Vision so many new keywords appeared.
When I scroll down to Metadata and look for a selection for Any Vision as shown on your screenshot ...there is none. I was just using their free trial. Could that feature be disabled in that?

 

[johnrellis]

"When I scroll down to Metadata and look for a selection for Any Vision as shown on your screenshot ...there is none."

Most likely the Any Vision plug-in is uninstalled or disabled in File > Plug-in Manager. Once it's installed and enabled, the Any Vision tagset will reappear, even if the free trial is expired.

 

[jjlad]

Ok ...Thanks John. I'll have to download Any Vision again and retry it. I don't think I even looked in that section of the metadata when I tried it before. Is it possible those numbers were there and have simply disappeared now due to uninstalling the app?

 

[johnrellis]

it possible those numbers were there and have simply disappeared now due to uninstalling the app?

The recognized text (OCR) is still in custom metadata fields in the catalog. When you reinstall the plugin and switch the Metadata panel to the Any Vision tagset, you should see the text.