Quality vs resolution vs size example

[Linwood]

Been experimenting to see what quality to use on the jpg captures, to see where the diminishing returns point is. It's not exactly easy.

Anyone know if there is any science behind this, e.g. for recognition of faces, based on resolution of the incoming image as well as the h.264 quality of that image?

I thought I would share an example for those interested. This is an extreme blowup (about 10x) of a 2048x1536 video image (medium quality setting on the feed). It is boat numbers on a boat at a dock about 40' or so from a 4mm camera image. Though I'm more interested in face recognition than reading numbers.

These are, top to bottom, quality of 50, 70 and 90. They take up space of 400k, 559k, 1015k respectively. So there's a large space penalty to get the additional quality.

I have yet to experiment with the video feed quality and bit rate settings, so there are at least two other dimensions to this question.



Obviously each person's solution may be different depending on disk capacity, network, etc. And I've read the general advice in the wiki and faq's, which is largely aimed at which things affect size, etc. But I'd love to know if someone has found a real sweet spot between these three (at least) variables.

[Southcross]

There is a double caveat to "compression" (compression, decompression, then compression again).. there is video/mjpg compression at the camera, then again at the ZM server. I lean toward very little compression coming from the camera, then a little compression at the ZM (since I am currently running SD resolution, I'm actually running NO compression LOL!). I get zero artifacting in my event files/images, and then with the suggestion I got from knnniggett (http://www.zoneminder.com/forums/viewto ... 32&t=23265) and very high bitrate settings for ffmpeg I get flawless video exports now.

Now there are also other factors that will effect image quality... clean your lens *giggle* and verify your focus, I made a 18x24 mounted and laminated target using: http://www.silverwinggraphics.com/focustarget.htm I get near perfect captures of faces and license plates now (well, as best as SD cameras will allow)

[Linwood]

Will look, but I tried yesterday setting the video (which is only h.264, no choices) to "highest" quality, and did not see any difference in the resulting images (at quality 90 from zoneminder). Nor any significant change in size (maybe 3-5%).

I don't want to check my focus, as the cameras are fixed focus and if I test it and find it is out of focus I will get depressed.

[Southcross]

Will look, but I tried yesterday setting the video (which is only h.264, no choices) to "highest" quality, and did not see any difference in the resulting images (at quality 90 from zoneminder). Nor any significant change in size (maybe 3-5%).

I don't want to check my focus, as the cameras are fixed focus and if I test it and find it is out of focus I will get depressed.

oi... fixed focus, and a fixed f-stop... you likely can't get much better than what you are currently seeing. I also have the added advantage of a auto-iris lens, when the lens stops down for bright light it increases depth of field. A fixed, wide-open lens, has a very narrow depth of field

[Linwood]

Cheap cameras I guess (about $100). I didn't really know what I was getting. They actually look better than I expected for a video capture. I've done image capture from very high quality gear (DSLR video playing around) and jpg capture from a video just generally is bad, in comparison to a still image. So I'm not sure how to compare the various shades of bad.

Of course it's much worse at night as it seems to increase the gain and noise. I need to finish the program I started to dynamically adjust the exposure parameters. I find a much slower shutter is a better compromise at night than the noise from the gain, but a faster shutter is obviously good in the daytime to reduce blur. Wide dynamic range changes are good also. The camera will switch exposure programs day/night, but it does so only on a schedule not based on light -- must have been done somewhere they don't have seasonal variations in day length. Another planet perhaps. But the good news is it is easy to reprogram with simple http requests.