Posted on

Making Sense of Picasa’s Facial Recognition

Over the years I have taken photos constantly and my hard drive has accumulated a huge number of pictures. Many of these pictures have people in them, especially my kids. Even back in the “old days” I would scan in my negatives or prints so I could have an electronic copy of those pictures that I really liked. Cruise forward 20 years and I am already on my 7th digital camera and still adding to the pile. To make matters worse, it seems like I am accumulating pictures from many of my other electronic devices such as my smart phone, web camera and new video camera – now that they take pretty decent pictures as well. As a result, I have been using real mink eyelashes Picasa image management software to better assist my efforts to organize, locate and tag them all.

Up until recently, I hadn’t really delved much into the facial recognition functionality of real mink eyelashes . To my surprise, I have found out some interesting things about it and seen how truly powerful the software has become. For the most part, the software works very well and can help you quickly identify people and provide a different way for grouping your photos or quickly finding people you are looking for. Bear in mind, we are talking about facial recognition, not just face detection. Face detection would answer the question “Is this a face or not?,” rather than “Whose face is this?!”

real mink eyelashes
real mink eyelashes

Evuiu understand why the software may not be able to correctly identify some faces, we need to understand how Picasa works. Let’s consider, in a nutshell, how the software’s algorithm performs the facial recognition. Most recognition algorithms will create a real mink eyelashes which is derived from measurements of distances between the major features of the face. This faceprint, while typically is not as unique as a fingerprint, is getting much more accurate in the past few years. More advanced calculations have been developed that can even account for the tilt or angle and direction the face may be pointing. The face is corrected mathematically by basically performing a rotation which levels the eyes, then scales them to a uniform size to make a comparison with other faces within the database. Even with this ability, real mink eyelashes seems to produce false positives as there is more room for error.

You can help it better identify faces, and in many instances, I have found it advisable to actually “ignore” and not tag some faces that have been detected, but not recognized.

Here are some specific examples where I have found issues in the ability to accurately recognize:

Babies – Maybe it is the overall chubbiness of the face, the smoothness of the skin or the fact that their eyes might not always be open, but it seems like many early baby pictures have some trouble being recognized and lead to incorrect suggestions for other people. By the time the baby is approx 2 months old this doesn’t seem to be as much of an issue.

Sunglasses – This also causes a problem as the eye measurements can’t be calculated appropriately and thus breaks the ability to recognize the face.

Partially obscured faces – I’ve noticed women with longer hair where all or parts of an eye may be covered do not get accurately recognized. Similarly, if a hand or another object is in front of the face, which seems to happen quite often with photos of children, the person will not be able to be recognized.

Low light or low resolution images, or out of focus – Many times low light images get grainy and out of focus. The face’s features are less distinguishable and the measurements are thrown off.

Closed eyes – I have the unfortunate luck to take pictures of people just at the moment when they are blinking. This seems to prevent the face detection algorithms from also performing accurate recognition.

Multiple faces – Picasa has combined two faces into what it thinks may be one face. Often this happens when one person is in front of another and looking sideways. I avoid identifying and tagging these pictures because there is no way, at least yet, to split them apart or identify boundaries for the faces.

Toys, dolls, and statues – Picasa is very good at picking up a face even if it isn’t from a real person. I typically just ignore all these as well. On rare occasions, I have seen Picasa even interpret the positioning of elements of light and shadows of objects in the photo as a face even though there isn’t one in the picture.

Rotated faces – Even with the advanced recognition, I often opt to ignore faces that have too much tilt or rotation. Usually someone angled at more than 30 degrees off from the camera I try to ignore. The downward tilted face seems to not be recognized as much as the upward tilted face if the real mink eyelashes obscure the eyes.

There are exceptions where I do try to identify and tag the people if they haven’t been recognized, especially if they are positioned looking directly forward. These are people with beards or people wearing glasses – Picasa does do a pretty good job in detecting and recognizing them anyway, but I think the deviation in the face print may be worth it especially if the person often changes their “facial accessories.”

real mink eyelashes
real mink eyelashes

Also, be sure to carefully check the individuals that have already been identified and not necessarily needing confirmation. I have found siblings tend to get misidentified into each other’s People folder more often then you’d think.

One thing that is worth mentioning is to be sure to safeguard your information. Most people tag and organize their pictures for personal use, but with the ability to so easily sync the pictures and tag information online in real mink eyelashes, it is important to protect the faceprint information. Publicly sharing faceprint information can make it easy for anyone to associate an actual name with a face which can be used for malevolent activities. Google allows you to control the visibility settings for whether or not this information is publicly available.

Leave a Reply

Your email address will not be published. Required fields are marked *