Work in progress...

Color Denoising

Color denoising algorithm

Wavelet shrinkage is a simple use of a wavelet transform in a denoising context. It gives intuition of how visual elements are encoded within this representation. We here show the advantage of the color definition over the classical use of channelwise processing, which processes red, green and blue channels independently.

Experiments show that the channelwise method produces false colors, while our elliptical monogenic wavelets prevents from this, thanks to an actual color model taking into account all the channels as a whole.

Results

Noise 20dB
Elliptical method : threshold is 0,51
Channelwise method : threshold is 0,38
Thresholds are fitted so that both methods keep the same amount of data.