I am trying to perform some post-processing of Camera Raw images captrued using my Canon EOS 450-D. I am using PS CS5 and GIMP particulary to remove noise which is present in images.

My questions is:

In a typical image post processing, I think that most of the denoising happens almost always in the last step of processing in YCbCr data space. I am looking at some method/tool to perform denoising in Raw bayer data space itself on the sensor bayer data. Can you suggest any such tools, or algorithms/papers which talk about this? (I intend to implement my own denoising of bayer data eventually if I find some algorithm for it.)


5 Answers 5


This is well beyond my understanding of the technology but I have read an article on this here. http://www.csie.ntu.edu.tw/~fuh/personal/NoiseReductioninRawDataDomain.A2-2-0103.pdf

It was an interesting read on the subject.

Also I notice you are using GIMP and CS5 for noise reduction, I found both to give me sub par results. I ended up settling on Nik Dfine for that process when I really need it and using Lightrooms built in noise filter as part of my process when I'm feeling lazy.


Denoising is a vast subject, there are a lot of methods and a lot of research papers about it in the literature.

A simple and efficient algorithm that you can take a look at is called Non Local Means. Basically it aims at averaging pixels that look similar as there are a lot of redundancy in an image. You can do it on the bayer (by taking into account the pixels of the same bayer color) or on a demosaiced image.

Have a look here: http://youtu.be/9tUns4HYtcw

You can look for articles and demo on http://www.ipol.im/ also, which is a great source of image processing features.


DxO 9 has a new PRIME algorithm that is marketed as being a very much improved noise reduction algorithm for high ISO images.

Here is a link to start you off on the PRIME algorithm. PRIME by DxO

I have used it myself and I am happy and quite impressed by the result. But it is very processor intensive and takes a long time to process each image.


Check out RawTherapee's LMMSE demosaicing algorithm- it's designed for noisy images, and offers options on noise reduction.


Image denoising is a big topic. Usually in the image sensor ISP, at least two denoise steps will be applied. One denoiser is for the raw image (maybe right after defect pixel correction), one is for the image preprocessed by lens shading, AWB, color correction and so on...

Non local mean is a simple and effective denoising method. BM3D is the state of the art denoising method.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.