MetaTagger

Facial Recognition

No, this program can not do any facial recognition.

But here is what I did!

By late 2021 I had a amassed a collection of ~27000 pictures. About a third were analogue: 35mm slides, 6x6 slides, 35mm negatives, 6x6 negatives and simple paper pictures
A pitiful small number of the pictures were old paper pictures of family members dating back to about 1908.
I felt motivated to preserve some knowledge (e.g. metadata) of the people and places in those pictures, and I wanted it to be in a computer file, not on the back of the paper picture.
A digital file, such as a database would allow me search, find and display the images and metadata.
The first step was to digitise the analogue images. So I bought:

• Epsom V600: flatbed scanner with picture / slide / negative capabilites (35mm and 6x6) er somewhat slowwww, but capable of very high resolution. Disappointing. Sorry, but it was not fun, it arrived with a failure in its lighting system, which was repaired, and then 13 months later failed again!
• Kodak Slide N Scan Film and Slide Scanner: a small 35mm slide and film strip scanner, reasonably fast, medium resolution. But it could not handle many of my slides because they where too thick. I solved this my making my own slide holder out of stiff paper and two thin strips of wood! It was fun but limited, I wish I had bought a similar scanner that also had 6x6 capability.
• Plustek ePhoto Z300: a quick and easy paper picture scanner. I found it fun to use! And it did what I wanted.

With these 3 scanners I was able to digitise all of my analogue images.
Even the panoramic picture from my university days which was too long to scan in one go. So I scanned as much as I could from each end, and then used a freeware program to stitch the two pieces together (I can not remember which program it was, sorry).

Now, there was one issue when digitising the images. Having dust free slides and film strips. It takes time!
My recommendation. Do not clean (yet)!
Scan everything. See which images are worth keeping, and which of these need cleaning.
Those that are interesting, clean them and scan to the highest resolution.

Now for the Facial recognition bit.

I tried using Microsoft Photos. Let us just say that I failed to achieve my goal!

I then heard that the Picasa program from Google, although no longer supported, has some facial recognition abilities.
Copies of Picasa can still be found in the Internet.
Downloaded it, installed it, ran it.
Within 2 hours I was applying names to facial groups that Picasa had identified.
Then another 2 days to sort out the problem faces (one baby can look a lot like another).

Of the ~27000 files, Picasa provided me with facial recognition on 7300 images.
About 5200 images with a single person
About 2100 images with more than one (identified)person

I then used the Region capabilites of this metadata tagging program and provided names by hand (easy, and with acceptable speed)
About 2300 images with a single person
About 700 images with more than one (identified) person

In total I now have 9500 images where someone is identified.

In some cases Picasa did not identify a face because the face was turned away from the camera, or was partially obsured.
All in all I am well pleased with the result.

Picasa saves, in plain text, facial recognition information in a .picasa.ini file which it writes in to the folder where the image is.
I have about 850 folders with images, so 850 .picasa.ini files.
I wrote a script of less than 300 lines, using the BASIC like language in which this application was written, to scan the .picasa.ini files, create a who tag for each face group(about 200 named face groups) and insert that tag in to the appropriate file's database entry.

Summary: facial recognition was much easier than I expected: Now I wonder if other recognition solutions are available (animals, flower,...) i.e. generic objects.

The .picasa.ini files include the position of the face on the image.
This was my motivation to implement Regions, see a separate help topic!