Like many others, I'm trying to decide between the Sony SR11/12 and the Canon HF100 and am interested to know if I should be factoring 'face recognition' into the decision?? I plan to use the cam to record mainly indoor and outdoor video of my kids and will be an 'automatic' user instead of using any type of manual control options.

I read the hype on the face recognition but would like to hear real-world feedback on its effectiveness before taking it into account to make a decision. From what I can tell, it will come in handy when shooting a scene where the subject is either in a shadow or light is coming from behind the subject. I know of several instances where I've had issues with this in the past (albeit with a simpke point-and-shoot digital camera) and if this feature works as advertised seems to be something I would want...even though I had pretty much decided on the Canon over the Sony.

Any advice on this would be appreciated!!

Here's my opinion and experience:

In automatic mode, it's definitely an amazing feature if you're trying to record people that one should weight in when deciding on a new camcorder. Not only the SR11/12's face recognition is helping with the focus (great by itself) but it also fine tunes the AE and white balance on the faces recognized. Having a family of diverse skin color, I was concerned about the white balance, but it's done really well, without skewing the white balance too much. IMHO, focus is the most helped with it, and the WB and AE skewing are more modest, definitely helping but not always completely correcting.

There are two other features related to it:

One is that the compression algorithm tries to give more bandwidth on the area where the detected faces are, thus reducing any compression artifacts on them. I took some pool videos of my children (notoriously hard to compress because of the entropy in the water), while I see artifacts on the water, their faces were unaffected. This is an appreciable improvement over my older AVCHD camcorder (the SR1), which I'm sure the face recognition helped in this case.

The second one is to be able to directly jump to any part of a clip where the faces were detected. The UI shows face thumbnails to let you navigate. I thought this would be a useless feature, until my daugther wanted to see herself on the clips I took. That helped a lot. :) That being said, I wouldn't call it a major feature...

One last thing I would like to note. I've had this camcorder for almost 2 months now and shot over 10 hours on multiple outings with it, and I've yet to see a false positive on the face detection recognition algorithm. If anything, the face recognition is a little too restrictive.

Is there any bad to it? A few issues IMHO...

1. Works as long as the faces are somewhat facing the camcorder in some way. If the subject is facing more than 60 degrees away from the camera, the face isn't detected and you don't get any of the avantages of this feature. Side profiles don't work a all. But note that you don't see a change in the focus, AE or WB when you lose track of a face.

2. Doesn't help on animals. I went to the zoo multiple times and the animal faces, even under the best conditions, don't show up (including monkeys). I was a little disappointed by that.

3. Any human face will be detected. That includes on TVs, screens, posters, books, etc. You may or may not want that.

ericjut,
that was a great post! I just wanted to say so.

I want to add my idea to what you mentioned about the face being turned 60 degrees and the detection not picking up the face. I kind of thought that the camera needed to clearly detect 2 eyes at least. Im not saying that the nose and mouth can be covered. Im saying a unobstructed face with both eyes visible to the camera. so what ever angle that is??

But this is just a total guess on my part. You have clearly taken much more notice to this than I have.

Again good post

Actually Paul, I think you're assesment is better than mine. It's most probably the 2 eyes it's trying to find. But it's also looking for other things, or else the Jaguar shot I took 2 weeks ago would certainly would have work too, since the beast was looking straight at me.

In any case, I would say it's really good feature to have if you're a point-n-shooter and/or running on automatic.

One is that the compression algorithm tries to give more bandwidth on the area where the detected faces are, thus reducing any compression artifacts on them. I took some pool videos of my children (notoriously hard to compress because of the entropy in the water), while I see artifacts on the water, their faces were unaffected.

I don't think it's because of face recognition. It's simply because water don't have consistency from one frame to the next, while your children still have the same face. If you film the borders of the pool, you'll see borders are unaffected too (consistent through frames).

I don't think it's because of face recognition. It's simply because water don't have consistency from one frame to the next, while your children still have the same face. If you film the borders of the pool, you'll see borders are unaffected too (consistent through frames).

They may still have the same face from frame to frame, but don't forget the water reflection on their faces, the movements (my daughter doesn't stay in place in a pool), the water dripping on their faces, etc. There's still plenty of entropy, in conjunction with the rest of the scene being water and light reflection on it, is stressing the compression algorithm to say the least.

You're most probably right though. The fact that the faces don't change must contribute a lot to a better quality on the faces. I am definitely seeing an appreciable improvement when compared to my SR1 footage in similar conditions, but it's not necessarly due to the face detection algorithm.

Independently of my experience, it's good to know that the "face detection" algorithm does affect the way the frames are getting encoded.

From the SR12's review at CCI:
"The purpose of the face detection, of course, is to make auto corrections based on a person’s presence. Sony states that exposure, focus, and white balance are all adjusted accordingly. It also states that the camcorder uses a new bit allocation system to direct more detail toward faces. We saw now direct evidence of this, though it would admittedly be difficult to test. We can only verify that the detection itself works very well. "

One funny effect of face detection is that it will look at wheels on a car as a face. When I did a sort by 'faces' in the scene index, I began to see wheels on cars being interpreted as faces! Pretty funny stuff.