minolta
15th May 2004, 05:11
Most people are still not 100% digital, and noisy analog capture still drives most of us crazy. A good way to get rid of noise is to average several samples together. For this test, a tv commercial was recorded four times from a bad-interference cable channel. Each recording taken alone appears extremely noisy, but blending the four together should look closer to the original.
In general, I don't mind allowing some noise to go through encoding. The problem occurs when that noise turns into ugly macro-blocks even at reasonably high bitrates. Compress-ability can be determined by using XviD with constant quality quantizer of say 2 (interlace enabled). The following results were obtained with the four sources:
Clip 1: 60.2MB (raw), 44.0MB (PeachSmoother)
Clip 2: 54.4MB (raw), 35.5MB (PeachSmoother)
Clip 3: 51.7MB (raw), 36.8MB (PeachSmoother)
Clip 4: 61.0MB (raw), 43.2MB (PeachSmoother)
To perform blending the following AviSynth script was created.
------
# Sources
c1=AviSource("c1.avi").Trim(5390, 6288).SeparateFields()
c2=AviSource("c2.avi").Trim(4196, 5094).SeparateFields()
c3=AviSource("c3.avi").Trim(645, 1543).SeparateFields()
c4=AviSource("c4.avi").Trim(1464, 2362).SeparateFields()
# Blend
b1=Overlay(c1, c2, mode="blend", opacity=0.50)
b2=Overlay(b1, c3, mode="blend", opacity=0.33)
b3=Overlay(b2, c4, mode="blend", opacity=0.25)
# Comparison (Screenshots)
#return StackVertical(c1, c2, c3, c4, b3)
#return StackVertical(b1, b2, b3)
Weave(b)
------
The following screenshot stacks the four clip sources and the blended result.
------
http://www.vrac.iastate.edu/~chadspen/cap1.jpg
------
Notice how smooth the bottom (blended) clip appears compared to the noisy sources. Indeed, blending multiple recordings seems to help quite a bit. Now I'm interested in how many recordings were truly needed.
Blend 1 & 2: 38.6MB (raw), 31.4MB (PeachSmoother)
Blend 1,2,3: 33.9MB (raw), 28.2MB (PeachSmoother)
Blend All : 30.5MB (raw), 25.9MB (PeachSmoother)
And the screenshot gives a visual comparison:
------
http://www.vrac.iastate.edu/~chadspen/cap2.jpg
------
So, my conclusions? 1) I must enjoy wasting time and 2) multiple recording, when possible, should be taken advantage of. Temporal and spacial filters all work great, but brute-force techniques sometimes work even better. Not sure how well the results apply to VHS recordings, but I imagine it can't hurt.
-Minolta
In general, I don't mind allowing some noise to go through encoding. The problem occurs when that noise turns into ugly macro-blocks even at reasonably high bitrates. Compress-ability can be determined by using XviD with constant quality quantizer of say 2 (interlace enabled). The following results were obtained with the four sources:
Clip 1: 60.2MB (raw), 44.0MB (PeachSmoother)
Clip 2: 54.4MB (raw), 35.5MB (PeachSmoother)
Clip 3: 51.7MB (raw), 36.8MB (PeachSmoother)
Clip 4: 61.0MB (raw), 43.2MB (PeachSmoother)
To perform blending the following AviSynth script was created.
------
# Sources
c1=AviSource("c1.avi").Trim(5390, 6288).SeparateFields()
c2=AviSource("c2.avi").Trim(4196, 5094).SeparateFields()
c3=AviSource("c3.avi").Trim(645, 1543).SeparateFields()
c4=AviSource("c4.avi").Trim(1464, 2362).SeparateFields()
# Blend
b1=Overlay(c1, c2, mode="blend", opacity=0.50)
b2=Overlay(b1, c3, mode="blend", opacity=0.33)
b3=Overlay(b2, c4, mode="blend", opacity=0.25)
# Comparison (Screenshots)
#return StackVertical(c1, c2, c3, c4, b3)
#return StackVertical(b1, b2, b3)
Weave(b)
------
The following screenshot stacks the four clip sources and the blended result.
------
http://www.vrac.iastate.edu/~chadspen/cap1.jpg
------
Notice how smooth the bottom (blended) clip appears compared to the noisy sources. Indeed, blending multiple recordings seems to help quite a bit. Now I'm interested in how many recordings were truly needed.
Blend 1 & 2: 38.6MB (raw), 31.4MB (PeachSmoother)
Blend 1,2,3: 33.9MB (raw), 28.2MB (PeachSmoother)
Blend All : 30.5MB (raw), 25.9MB (PeachSmoother)
And the screenshot gives a visual comparison:
------
http://www.vrac.iastate.edu/~chadspen/cap2.jpg
------
So, my conclusions? 1) I must enjoy wasting time and 2) multiple recording, when possible, should be taken advantage of. Temporal and spacial filters all work great, but brute-force techniques sometimes work even better. Not sure how well the results apply to VHS recordings, but I imagine it can't hurt.
-Minolta