Waiting for the exact moment to snap a picture can be very difficult especially when working with children or animals. I've read somewhere (can't find source) that every freeze frame from an HD video is essentially a high-resolution photo (is this even true?).
What are the challenges, pros, and cons of selecting the best frame from a video as a way of taking still photos where timing is crucial?
The big pro to this is, as you say, the ability to take many pictures in a short period of time, allowing you to pick the best frame.
However, there are several cons to this approach:
I've read somewhere (can't find source) that every freeze frame from an HD video is essentially a high-resolution photo (is this even true?).
No, not true at all really :( Here's why:
First up, it's not high resolution. Even 1080P ("Full HD") video, which not all cameras can do, is only 2 megapixels. That's a fraction of the sensor's resolution. 720P video is 1 megapixel.
Secondly, video uses a much higher compression rate than still JPEG images, because video is bandwidth-intensive. This means the resulting video frames are (generally) more compressed than an equivalent still image would be. Additionally, in most video formats (including AVCHD, H.264/AVC, various MPEG/DivX formats, but not "Motion JPEG"), most video frames use "inter-frame compression" which means they are not stored as images in their own right but as transformation data to apply to other recent or nearby frames.
In video, a certain amount of motion blur is desireable, but in stills you would not want nearly as much. 1/48 to 1/60 second is normal for 24P and 30P in order to have a natural feeling motion blur (the effect is called "shutter angle" after old movie cameras). Otherwise the motion can look unnatural. In still images, freezing motion using a shutter speed of 1/250s or faster looks fine, and is often desirable.
Digital sensors often can't provide such good quality image when they are in "Live view" mode, where there is no physical shutter but instead they're continuously fetching frames from the sensor. In CCDs this manifests as vertical streaking; in CMOS there are other issues (one being waviness during camera motion, known as "rolling shutter effect"). The shutter that operates when taking stills enables the sensor to use a different mode where it doesn't have to reset so quickly and it can take a high quality image with less noise.
Video often has additional filtering applied than a still would, which loses some detail. One major one is interlacing, which effectively throws away a lot of vertical resolution. However, not many stills cameras with a video function would support interlacing. A lot of video will still have an antialiasing filter applied, which softens the image but can help with compression and reducing moire. Depending on camera, video may also have more sharpening applied although nowadays with HD this practice is not so prevalent/detrimental.
What are the challenges, pros, and cons of selecting the best frame from a video as a way of taking still photos where timing is crucial?
Unless you had no other method whatsoever, I wouldn't recommend it. In fact I probably still wouldn't recommend it even if it was your only option. But if you did go ahead with this then make sure you can do 1080P, set the shutter speed to fast enough as if you were shooting still, and use Motion JPEG format (not AVC or any of the other formats I mentioned above). And realise that you're getting basically 2 megapixels.
Edit: Since writing this answer some time ago, I am no longer convinced that Motion JPEG mode (if your camera even supports it) is a good idea for this. Your camera would need a much higher bitrate for the same quality. If you could use regular HD video (eg. h.264 or AVCHD, etc) and find some way to find the I-frames, that might be better. I still don't think it is a good idea overall though. Modern DSLRs support much faster continuous still shooting modes including some that can lock the mirror up or decrease resolution and shoot faster, etc.
I'll use my own camera (Pentax K-5) as an example (width x height):
So, I don't think it's a high resolution photo. Having said that, getting the shot at any resolution is better than not getting it at all. :)
Take a look at the Red cameras, if this is something that you really want to try out. Their whole raison d'etre is to shoot at something like 5x HD, and each still would be a 10 megapixel shot.
Of course, they aren't cheap, so it might be a better idea to get a camera capable of shooting, say, 8 fps for a few seconds and combing through those photos for that perfect moment. I've found that 5-6 fps for 3 seconds is usually more than enough time to get a good shot, or at least, provides the maximum number of frames I want to look through per burst. If you go full video rate, that is a lot of data to comb through for just one still.
This should have been a comment to the post by @thomasrutter, but I can't comment yet (lacking reputation).
Your bullet point about MJPEG versus AVC is not entirely correct. It is true that MPEG-4 Part 10 (AVC /H.264) usually doesn't record every frame as an I-frame (key / full frame). There is nothing in the specification that prohibits this though, and some cameras do record only I-frames.
What you are wrong about is the benefits of MJPEG over H.264, and your argument about why. I'll start with a short explanation about how the frames are reconstructed from a H.264 (or other video codec formats). The video will be coded in three different types of frames, intra-coded frames (I-frames), predictive-coded frames (P-frames), and bidirectionally-predictive-coded frames (B-frames). An I-frame is a compress version of a full "raw" frame. It can be reconstructed just as a JPEG (or a single frame in a MJPEG stream). A P-frame on the other hand reference preceding I- and P-frames, and will need these to be reconstructed. A B-frame can also reference following I- and P-frames. It's a lot more information about this on Wikipedia.
The difference between H.264 and MJPEG is how the compression works, and not how good it is. Every single frame in a digital video can be completely reconstructed, by using information that is in the video-file/-stream. H.264 will give better results than MJPEG, even when you export one frame, at the same bitrate. This is valid for most bitrates (not sure about very low bitrates). There is more information and several tests on line, and a good place to start is compression.ru.
The only disadvantage with H.264 is the computational complexity in reconstruct a single frame. H.264 requires a lot more processing power than MJPEG. This is not a problem on a modern computers though.
