Using Facial Recognition to Compare Photos and Identify Individuals

December 20, 2020

By now, you’re probably aware of the facial recognition built in to many photo storage programs. When you upload a new photo, a program like Google will automatically try to identify who is in the photo, based on labels you have previously created from other photos you have stored. Every now and then you will have to correct a label or two, but I find it to be highly accurate.

It may not have occurred to you that you can use these programs to identify individuals in your stack of old ancestor photos - if you have at least one other photo to compare with, that is confirmed to be the individual in question.

How many of us have old photos we know must be ONE of our ancestors, but we can’t figure out who? I see many posts across various Facebook genealogy groups, asking if two different pictures could be the same person, so I know I’m not alone in this!

Programs that currently have facial recognition for public use:

After uploading to Google Photos, you may have to wait a bit for the facial recognition to kick in (up to 24 hours) and be sure that the feature for “face grouping” is turned on in your settings. You will see the results grouped in a “People” album, without names unless they match someone already given a name in your photos. The accuracy is pretty good, but not 100%.

There are also online reverse image searches, in which you upload a photo or a link to a photo, and search for similar images online. Some, like Google Images, are not based on facial recognition but on similarities within whole images. Others, like PimEyes, actually use facial recognition technology - but it might cost you.

In addition to these programs, there are many additional programs, some free with registration, that offer facial comparison between photos, that I have not had the opportunity to try yet - except for the demos that is. And I love the demos!

Above: A screen capture of the Face++ demo which I used to compare a young soldier and an older gentleman. I knew the soldier was one of a few brothers, but not sure which. Using facial recognition I was able to get a “probability high” result on one of them. For further details, you can click the second tab “Response JSON” while using the demo to see the code. I am able to learn the confidence level is 71.6% positive.

With privacy concerns in mind, I’m still crossing my fingers for some more robust genealogy tools in the near future. How amazing would it be to instantly identify all of our old photos - or find a wealth of new-to-us photos of our ancestors??

Have you had success using these programs to identify old photos? Or know of any other similar tools? We would love to hear from you in the comments.

When Abstracts are Wrong

Dec 1, 2020

My awesome cousin brought this example to me, and it’s a great one to show how abstracts can have misinformation. With abstracts, information is extracted and summarized (usually by a computer algorithm). You may see them on database sites like FamilySearch.org or Ancestry.com when you click on a promising record like a census. The abstract will show you the names of the head of house and their family members, ages, occupations, and so on.

This is similar to transcriptions in which information is copied, usually by hand, to the best representation possible. Both abstracts and transcriptions allow room for errors and should not be given as much weight as the original source (see Types of Sources).

An abstract of an 1861 Census of Canada courtesy of Ancestry.com

An abstract of an 1861 Census of Canada courtesy of Ancestry.com

A first glance at the abstract looks like our subject, Judith Cook (misspelled in the census as “Judath”) is Native American. Let’s open the census and examine closer.

Above: selection from the 1861 Census of Canada, New Brunswick, Westmorland, courtesy of Ancestry.com, with “native” circled in red.

Above: selection from the 1861 Census of Canada, New Brunswick, Westmorland, courtesy of Ancestry.com, with “native” circled in red.

The census shows Judath Cook (or rather “do” as short for “ditto” meaning the same as above, which is “Cook” in this scenario) below her husband, Samuel Cook. The header is out of sight in this selection, but the column to the right reads “Race and where born”. Here it says “Nova Scotia” for Samuel Cook, and “native” for his wife.

We can assume that one of Ancestry.com’s algorithms picked up “native” in a column for race and calculated her ethnicity to be Native American in the abstract. However, if the answer for her husband is related to the place (Nova Scotia), it’s more likely that in this case, native means native to the area, as in born in New Brunswick.

We can confirm this by looking up information about the 1861 Census of Canada. On the Library and Archives Canada page for this census we find the following disclaimer: “In the 1861 census of New Brunswick, “Native” meant born in that province, not Aboriginal.”

We can now come to the logical conclusion that the abstract is incorrect, Judith Cook’s race is undetermined from this census, and that she is native to New Brunswick.

Other sources may show her race, and we may come to discover later that she is Native American, but this specific source does not show this.

Long story short, be careful with abstracts!



Sources: