Face recognition and keywords

[ERBRO]

Hello,

I'm testing the trial version of iMatch as I'm looking to replace my actual DAM.

In the past, I stored the people contained in a picture as keywords.  I discovered the face recognition embedded in iMatch. 

1) I can not figure out where the persons assigned to a picture by the face recognition tool stored. Is it only in the iMatch Database or somewhere in a metadata field or XMP file ? I use capture one and can not find the information in the metadata.

2) IS there a way to copy the persons assigned to a picture by the face recognition tool to the keyword tag ?

Any advice in this area is welcome.

Best regards
Eric

[Mario]

Hi, welcome to the community.

1. IMatch stores face and person data in the database, like all other metadata.
When you write-back metadata (see Metadata Write-back), IMatch stores the face regions (green rectangles) in the official XMP face region metadata, and assigns the person tag to the region. This makes the faces available in all other XMP-enabled software, if it supports XMP face regions.

2. You can associate keywords with a person (See Person Keywords and Categories) and when IMatch (or you) assign a person to a face, the keywords associated with that person are automatically added. Same for IMatch categories.

When you remove a face region or change the assigned person for a face, the corresponding keywords are automatically removed or updated.

This allows for a very efficient workflow. IMatch or you assign persons to faces, which adds keywords, which automatically organizes the persons in the @Keywords Category. And then there is the People View, which organizes your images based on the persons they show.

When you write-back, IMatch stores the keywords in the official XMP keyword tags, making them available for other software.

If you are just starting with IMatch, remind yourself that IMatch is a very powerful DAM software, with tons of features you have probably never seen or used before Feel free to read interesting sections in the IMatch help system (just press <F1> anywhere in IMatch) and post a question here, I you need help or advice.

For all things related to face recognition and the closely related "people" features, I can only recommend to spend an hour to read the People help topic (or at least the sections which sound interesting to you). This will show you how powerful these features are and will give you ideas for how to manage your images better and quicker.

There are also related free tutorial videos in the IMatch Learning Center.

[ERBRO]

Hello Mario,

thank you for your quick feedback.

I still try to figure out what is the best option. 

In the past, I added persons as keywords (> 50 000 pictures).   If for the new ones I start to using face recognition to speed upthe process, I will end up with a mixed of persons identified as keywords for the old pictures and persons identify by the face recongnition for the new.

I see as options

1. Run the face recognition on all old pictures what looks to me quite time consuming
2. Copy in an automatic way, the face recognition results of the new pictures to keywords. But I did not find a way to do that

Are the any other options to cope with this mixed situation and to get a consistency for the search (either via face recognition or via keywords) without having to spend to much time on it ?

Best regards
Eric

[Mario]

If for the new ones I start to using face recognition to speed upthe process, I will end up with a mixed of persons identified as keywords for the old pictures and persons identify by the face recongnition for the new.

Why? If you use the same keywords for persons in IMatch you have manually assigned before, there should be no difference?
I mean, if you assign "Paul" or "Person|Paul" manually or let IMatch do it, the result is the same. Or not? In that case we would need more details from you.

1. 50K images is not that much, Give it a test on 1,000 images to see how fast it is on your PC.

2. Why?
IMatch detects a face and assigns the best matching person. You accept that or change the person.
The keywords of the assigned person are assigned to the image. There is no need to copy anything in an automatic way, or manually.

[Tveloso]

2. Copy in an automatic way, the face recognition results of the new pictures to keywords. But I did not find a way to do that

IMatch is actually already doing that.

As Mario says, if you assign the same keywords you have been using, to the Person Records you create, then IMatch will add those same Keywords automatically to the files you run Face Recognition on.  So your new files (which will contain both Faces and Keywords), will be no different from the old ones, from a Keyword perspective.

It's the old files (which contain only Keywords), that will differ from the new.

1. Run the face recognition on all old pictures what looks to me quite time consuming

Another option for doing this quickly might be to use Face Links.

You could go to the @Keywords Category for each Person, select all files in the File Window (Ctrl+G), and then press Ctrl+M,L, to assign that person to all files at once.  IMatch will not duplicate the Keyword for the Person (which these files already contain).  The only concern might be if some of the files you  have selected for this, are new ones (which already contain true Faces)...I'm not sure how IMatch will handle a request to add a Face Link to a file that contains a true Face Annotation for the same person.

This is still a bit of manual work (because you will have to do this for each Person Keyword), but maybe a little faster than running Face Recognition on all your old files.

But that said, it would probably be better to use true Faces (i.e. runFace Recognition), than to use Face Links.

[Mario]

But that said, it would probably be better to use true Faces (i.e. runFace Recognition), than to use Face Links.

That^

Face links is a convenient band-aid/patch if you want to associate persons with images, but you don't want or can't run face recognition. Which should be rare.

A link does many things a real face annotation does, but the faces in the image don't partake in future face recognition, cannot be trained and there is no face thumbnail or spacial data about where in the image the faces are placed or how large they are in relation to the image. Which prevents links from being used in features like Face Arrangement.

In my case, when IMatch fails to detect a face, I preferably place a manual face annotation and assign the face.
I find rare use cases for face links, except for an initial, quick, pass to link images to faces.
Other users might do this differently or using face links more often.
IMatch gives users options, for that reason.