# Noise modeling and reduction in Video when we have just noisy video

We would like to send a video (sequence of images), we have limit bandwidth. If there is a lot of noise, the required bandwidth (bitrate) is needed to send the video+noise becomes high that is not suitable. We have just a noisy file (i.e. we do not have noise separately). The solution is to model the noise. we do not have any other information rather than a noisy video. How we can reduce the noise? How we can model the noise?

• Can you add some more description (screenshot) to see what type of noise it is really? Can you tell us what is the source of noise? Is this due to poor capture, loss due to analog transmission or MPEG packet drop? is this grainy, blurry what is it? Jun 13, 2015 at 8:27

Usualy, when you send a video over a network, the video is compressed. For instance, let's say your video is compressed using a wavelet transform. The most simple way to compress your video is to truncate the wavelet transform 's coefficients. Apply a wavelet transform on your noisy video, if your noise isn't a simple white noise, it will have a specific template. Then you can use this template to tune your initial wavelet transform 's truncation.

Strictly speaking about noise modeling, without going for any transform, you could also take all your noisy video, and build the noise's pdf in the grey level scale (one dimension histogram). If your space is RGB (3 dimension histogram), you'll prefer to use a kernel density estimator. You could also take some spatio-temporal patches say 5 temporal frame, and 50*50 spatial patches, you'll have a lot of patch with your video, treat them as realization of a random vector, so consider them as a vector of length 5*50*50, you can now calculate you noise covariance matrix, which is another modelisation. It will allow you to go for denoising/compressing with PCA.

imnoise(video,noisename);