Security cameras, just like many other sensors that gather data from the outside environment, are not invulnerable to noise. The footage captured by a security camera can have errors and corrupted information because it couldn’t get rid of the noise. (or collect more data to override it.)
Making sure that a camera captures footage without noise is difficult to control as it depends on the sensor. However, experts have figured out several techniques to reduce noise. These are called “Digital Noise Reduction” techniques and that’s what we are going to discuss in this article.
What Exactly is Noise
In the simplest of terms, noise is unwanted information. In cameras, noise affects the footage, and in some worst cases, can destroy vital information in a frame. (In our case, a frame is an image captured from a camera.)
Noise will always be present in an image – because sensors are not perfect devices. However, it only becomes apparent during low-light conditions. So if you take a photo in a dimly-lit environment, you might see that the image (or some regions of it) is slightly grainy with small pixels/group of pixels with color changes. These are the visual representations of noise in an image.
The reason for the occurrence of noise is because of drawbacks in the camera’s light sensor. Although cameras have come a long way, their basic function relies on collecting enough light from the environment. Therefore, the more light it can collect, the better the image will be. However, since security cameras need to collect footage in real-time (ideally 24 or 30 frames a second) we cannot allocate more time for the sensor to form a single frame. Therefore, in low-light conditions, the camera will have a certain amount of noise present in the image.
What is 2D DNR?
2D Digital Noise Reduction is one the most basic methods used to eliminate noise. Although it is successful in getting rid of noise in images, it doesn’t do a great job in higher resolutions and when there’s a lot of motion around.
2D DNR is considered a “Temporal Noise Reduction” technique. What happens is that each pixel on each frame is compared to the pixels on the other frames. By comparing the intensity values and colors of each of these pixels, it is possible to develop algorithms to detect a pattern that can be categorized as “noise.”
After these unwanted patterns are discovered, the system will try to clear out the noise by assigning an intensity level/color to a pixel that it deems as noise-free. Although this might work for a static environment where there’s no movement, it can cause motion blur as the system tries to do its best to figure out which color/intensity is the most natural for a certain pixel. Also, depending on the computing capabilities of the system, it isn’t possible to process these algorithms in larger resolutions and provide them in real-time. (They can process, but it takes a lot longer.)
What is 3D DNR?
3D DNR is quite different. It is considered as a “Spatial Noise Reduction” technique, (although it does have minor temporal aspects to it) where it compares and processes pixels in three dimensions. (First the two dimensions in an image, and then the third dimension, when it compares pixels with the other frames.)
Due to the difference in the algorithms, 3D DNR can be processed much faster than 2D DNR. Therefore, this algorithm can be performed in frames with a larger resolution.
During the 3D DNR process, the system compares the pixels in a single frame. It tries to eliminate noise within a single frame using techniques such as “median filters” and “averaging filters” to eliminate the grainy nature of a single frame. After that, it compares the pixels on a frame-by-frame basis to pick up any motion in the footage. (Motion is picked up by considering the changes in brightness/color on the same pixel coordinate in two (or more) different frames.)
Impact of Brightness on Image Noise
As we mentioned previously, noise is more apparent in images when the footage is captured in a dimly-lit environment. Therefore, we can safely conclude that the brightness of a scene can affect the apparent noise in a frame/image.
Cameras, like many other sensors, are not perfect, they are vulnerable to noise and the measurements that they make will always be an approximation of the real world. Therefore, every image captured by a camera will have some form of noise and it is a battle of gathering more information to overcome the noise i.e, getting a higher SNR. (Signal to noise ratio.)
Getting more signals to overcome the noise is where brightness comes into play. The brighter and more well-lit an environment is, the better the camera is at collecting light to produce a clearer image. So when a scene is captured in low light, the camera doesn’t have enough information to go on and this causes noise to accumulate.
So even though post-processing techniques, such as 2D and 3D DNR, can reduce the noise to some extent, improving the lighting conditions before capturing footage is always the better option.
Impacts on Storage
Noise doesn’t affect the storage directly. However, it can affect the file size if the user tries to change the resolution of the image/video files.
The reason for this is, because of the complexity of the DNR algorithms, some cameras and systems do not have the processing power to compute these algorithms in real-time with minimal latency. Therefore, to make the noise less apparent, some systems might opt to lower the resolution of a frame/image.
Lowering the resolution means fewer pixels to process. Hence, this allows the system to compute the DNR algorithms faster and take up less storage space on the hard drives.
Therefore, noise doesn’t directly impact the storage, only if the user/system deems it necessary to adjust the resolution of the frames/images.
Impact on Motion Detection
Noise can affect motion detection algorithms. As we mentioned previously, the 2D DNR algorithm does not take kindly to constant motion in the footage as its post-processing algorithm causes motion blur in the final footage.
Fortunately, motion blur does not hinder motion detection. The system will always pick up motion as long as the motion blur effect does not blend in with the environment. However, motion blur does have its downsides, the blurry nature of the object in the frame will make it difficult to distinguish important information such as faces and number plates. Therefore, for motion detection and other requirements of gathering information, it’s always best to minimize the motion blur effect either by using 3D DNR methods or similar post-processing techniques.
Temporal Noise Reduction vs Spatial Noise Reduction
From what we’ve gathered so far, we’ve determined that 2D DNR is a temporal-based method while 3D DNR is a Spatial-Temporal combination with the primary algorithms performed in the spatial domain.
Although both methods can eliminate noise, they do have their own set of drawbacks. The 2D DNR (or temporal noise reduction method) is not very complicated to set up. However, it takes an unacceptable amount of time to process larger images and happens to leave a motion blur effect after it’s done. It also requires multiple frames for noise elimination and cannot be used to denoise a single image.
Spatial Noise Reduction techniques are complicated to set up and require systems with advanced processing power. However, their algorithms are faster and work seamlessly with cameras that have larger megapixels. Also, they can work with just one frame and if necessary, can denoise multiple frames without leaving a motion blur effect.
Benefits of These Features For Camera Surveillance
Noise in security camera footage can become very irritating. They can make it harder to distinguish what’s going on, and can also affect high-level processing algorithms such as motion detection, face recognition, number plate recognition, etc…
Also, if the system doesn’t process the noise and optimize frames – after denoising, the images can be downsized, since the images are much easier for humans and machines to interpret – it can cause a burden on the storage system because of the accumulating RAW files.
In the same way that prevention is better than cure, adjusting the lighting of the environment is always recommended to eliminate noise. However, it’s not always under our control, and therefore, having digital noise reduction techniques always working on the back-end has become something of a necessity.
Digital Noise Reduction techniques are important for CCTV security cameras because they help eliminate the grainy effect caused by capturing footage in low-light conditions. Eliminating noise helps improve the clarity of the image, allowing for better human/machine interpretation. Most CCTV cameras use two types of denoising algorithms, the 2D DNR and the 3D DNR. While 2D DNR has a simple approach and can be found on many older and inexpensive systems, the 3D DNR is the more favored algorithm because of its ability to denoise larger resolutions without causing motion blur.