Method related to noise reduction
View Patent ↗A pre-processing of video data before coding and transmission, for a method for noise reduction in High-Density video images. A low implementation-complex filter filters raw pixel data output of a video camera with a low pass property for both spatial and temporal noise and the filter is adapted based on the high frequency content of the picture, leaving most of the image content unaffected by the filtering process.
1. A method of noise reduction of a video image being composed of succeeding pictures, the method comprising:
consecutively calculating a respective delta value (Δ) for each pixel in a present picture of the succeeding pictures from a function of weighted pixels in said present picture and in a reference picture relative to said present picture;
adjusting said respective delta value to a corresponding respective adjusted delta value (Δ′) according to an amount of high frequency content in said present picture and in said reference picture; and
adding said respective adjusted delta value for each respective pixel in said present picture, the consecutively calculating said delta value comprising:
providing a first version of said function of weighted pixels, weighted towards temporal filtering,
providing a second version of said function of weighted pixels, weighted towards spatial filtering, and
selecting the version to use in dependency on the difference between each version of the delta value and the pixel value in the same pixel position in the former picture.
2. A method according to claim 1 , wherein said second version of said function of weighted pixels, weighted towards temporal filtering and calculating said respective delta value for a pixel (e p ) in the present picture, is defined as:
Δ=(4 e r +2 f r +g r +2 h r +i r +a p +2 b p +c p +2 d p )/16− e p ,
wherein the weighted pixels are positioned adjacent to e p and weighted pixels with p-subscripts are included in the present picture and weighted pixels with r-subscripts are weighted pixels from the reference picture.
3. A method according to claim 1 , wherein said first version of said function of weighted pixels, weighted towards temporal filtering and calculating said respective delta value for a pixel (e p ) in the present picture, is defined as:
Δ=2 e r +( a p +c p +1)>>1+2[( b p +d p +1)>>1]+( g r +i r +1)>>1+2[( f r +h r +1)>>1]−8 e p
wherein the weighted pixels are positioned adjacent to e p and weighted pixels with p-subscripts are included in the present picture and weighted pixels with r-subscripts are weighted pixels from the reference picture, and >>1 is defined as a right shift by 1 in a shift register.
4. A method according to claim 1 , wherein said function of weighted pixels calculating said respective delta value for a first pixel (e p ) in the present picture is defined as:
Δ= w*Δ s +(1− w )*Δ t ,
wherein w is a value between 0 and 1 being proportional to |e p −e r |, and e r is a second pixel in the reference picture positioned equally as said first pixel, Δ s is defined as:
Δ s =(4 e r +2 f p +g p +2 h p +i p +a p +2 b p +c p +2 d p )/16− e p
and Δ t is defined as:
Δ t =(4 e r +2 f r +g r +2 h r +i t +a p +2 b p +c p +2 d p )/16 −e p .
5. A method according to claim 1 , wherein said respective adjusted delta value is defined as:
Δ′=Δ*max(0;( k 0 −k 1 *|Δ|)/ k 0 ),
wherein k 0 and k 1 , are constants.
6. A method according to claim 1 , wherein the video image is a video image captured by a camera used for video conferencing.