Digital Video Processing with HD CCTV: Noise reduction
Digital Video Processing with HD CCTV: Noise reductionIn this post I examine some of the sophisticated technology used in digital high definition CCTV cameras, including a the powerful underlying SOC electronics, and the similarities with audio noise reduction in audio, as used in amateur radio.
Noise reduction in HD CCTV
I read very widely and like to see or create connections between apparently disparate ideas and technologies. From an earlier post on working with a Pentax lens I have, I read about the image processing abilities of some Pentax lenses.
Pentax/Richo have an award-winning image processing system, PENTAX Atmospheric Interference Reduction (PAIR). http://www.ricoh.com/fa_security/security/pair/ The results are amazing:
From a first reading, the processing seems to be done in the lens itself, something that had me mystified as to how such spectacular results are possible with just optics. However, on closer reading, the output of the camera is fed into the lens and the main output is from the lens. As such, the lens has on-board processing of the captured images, making such image processing more plausible, compared to doing it with optics as I first thought. The connection are shown:
The lenses are expensive, for CCTV lenses, at about $10,000. Having realised the PAIR system uses the raw images from the camera, then it is easier to see how it might work; it is very difficult to find technical information on the PAIR system, secrecy can be better than patents for protecting intellectual property, something I have experienced myself during my career with technology transfer.
On the other hand, Texas Instruments, who make a very sophisticated system on a chip (SOC), video processing system (VSP), has an interest in telling the world of its possibilities, in order to sell more chips; even a Wiki: http://processors.wiki.ti.com/index.php/Category:DaVinci.
The TI Da Vinci SOC is essentially a video camera on a chip plus powerful VSP. The block diagram of a "reference design" networked camera: http://www.ti.com/lit/wp/spry224/spry224.pdf
With this SOC, it is possible to produce very powerful CCTV cameras at low cost. I purchased such a camera, with a 2K UHD sensor, but no lens, for about $170 from http://www.aliexpress.com/. It was purchased to investigate DATV over networks and UHD DATV, subjects of earlier posts. However, with the DSP, there are other possibilities. Taking the lid off the camera revealed a very simple construction. The main SOC is obscured by a heatsink, while the other chips were memory or auxiliary.
Low-light noise is a bug-bear of CCTV security systems. Using the TI SOC, noise reduction is possible: http://www.ti.com/lit/wp/spry224/spry224.pdf (Nov 2012, so close to state of the art)
Looks familiar to Pentax PAIR technology? TI's solution is to put the DSP in the camera and use conventional lenses, whereas Pentax's solution is to put the DSP in the lens. TI go on to discuss the algorithms.
Amateurs are fairly familiar with noise, with much of it being random, likewise with video. TI identify two techniques to remove noise: spatial and temporal filtering (see pdf for details). The essential approach is to compare a number of frames (typically collecting 25 or 30 frames per second), with little motion. What is in one frame, but not another is usually noise and can be removed. This is very similar to noise-cancelling with audio and radio where two different sources (two antenna, two microphones) are used and the differences discarded as noise.
Temporal filtering is more complex as it is used with moving images and needs to differentiate noise between frames as well as motion between frames; a two level system with different time regimes.