Dump Those Silly Colored 3D Glasses


# AnaExtract.avs
# Tone at VRtifacts.com
# V 0.9 June 27, 2010
#
# Copyright (C) 2010 Tony Asch
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>

The main points are: There’s no warranty, feel free to modify and distribute, but you must include the source in any distribution.

# Snag our video files
vidL = DirectShowSource(anaglyphName)
vidOrig = DirectShowSource(anaglyphName)
vidsound = DirectShowSource(SoundName, audio=true, video=false)


The parameter AVS file identifies the video file paths and names. Here we open a couple of copies of the anaglyph movie and a copy of the audio only.


# We need these for later
# Skip if we've already got a Mono Left source
# Red/Cyan?
(inputFormat == "RC") && (isMono != "monoLeft") ? Eval("""
vidL = ConvertToRGB(vidL)
vidL = MergeRGB(vidL.ShowRed("YV12"), vidL.ShowRed("YV12"), vidL.ShowRed("YV12"))
vidL = Greyscale(vidL)
""") : Eval(""" """)
# Green/Magenta?
(inputFormat == "GM") && (isMono != "monoLeft") ? Eval("""
vidL = ConvertToRGB(vidL)
vidL = MergeRGB(vidL.ShowGreen("YV12"), vidL.ShowGreen("YV12"), vidL.ShowGreen("YV12"))
vidL = Greyscale(vidL)
""") : Eval(""" """)

# Skip if we've already got a Mono Right source
# Red/Cyan?
(inputFormat == "RC") && (isMono != "monoRight") ? Eval("""
vidR = ConvertToRGB(vidOrig)
vidR = MergeRGB(vidR.ShowGreen("YV12"), vidR.ShowGreen("YV12"), vidR.ShowGreen("YV12"))
vidR = Greyscale(vidR)
""") : Eval(""" """)
# Green/Magenta?
(inputFormat == "GM") && (isMono != "monoRight") ? Eval("""
vidR = ConvertToRGB(vidOrig)
vidR = MergeRGB(vidR.ShowRed("YV12"), vidR.ShowRed("YV12"), vidR.ShowBlue("YV12"))
vidR = Greyscale(vidR)
""") : Eval(""" """)

Since many 3D DVDs come with both a 2D and a 3D copy of the same movie, it is often the case that the 2D version is a perfect reproduction of either the left or right eye view. If that is the case, we’ve cut our work in half, and are assured of one eye’s view being perfect. The block of code above separates the two anaglyph colors and produces independent left and right video streams, albeit that it’s primarily luminance information that we get once they are transformed to greyscale. If we already have something from the 2D copy, then we don’t need to extract an imperfect copy of that eye’s view from the anaglyph. Cool! In the case of RC, I found the blue channel to be quite noisy and even though cyan is a combination of blue and green, I just grabbed the green and used it as luminance information. If you don’t have a 2D version you will set the parameter isMono = “monoNone”, the parameter vidOrig will be set to “nothing” or some other string, and will not be used by any of the code.

# Prepare color information by resizing the image down and back up,
# creating a blurred version of the source for color restoration
# Note: the decimateHoriz/Vert values are percentages (1-100)
# defining the size of the shrunken version. Small numbers give more blur
# Code has a hack to make sure the shrunk version dimensions are even numbers
vidColor = DirectShowSource(PureColName)
vidColor = BicubicResize(vidColor, Int((width(vidL) * decimateHoriz) / 200.0) * 2, Int((height(vidL) * decimateVert) / 200.0) * 2)
vidColor = BicubicResize(vidColor, width(vidL), height(vidL))

In the prior step we extracted greyscale copies of the left and right video streams. In preparation to re-coloring them, we need to calculate the colors to apply. Since the left and right videos are very similar, perhaps offset horizontally in places, and knowing that the human eye perceives much more detail in luminance than it does in chroma, we can apply color with a very blurred paint brush. Back in the parameter .AVS file we defined the name and path of the best video file to use for extracting color information. This would be the 2D version, if we have it; otherwise it will be the anaglyph version.

In AVISynth, the quickest way to blur is to resize a video to a very small size and then resize it back up to full size. For the convenience of users, the parameter file specifies this independently for the width and height, both in percentage terms (1-100.) The code with ” * 200 / 2″ scales from percentage to fraction and also assures that width and height will be an even number (required by AVISynth.)

Horizontal and Vertical color blurring are specified independently because anaglyph 3D images are offset primarily along the horizontal dimension and we need much more blurring horizontally than vertically. Nonetheless, many movies are shot with imprecise 3D cameras which are not perfectly aligned and may exhibit dissimilar lens distortions between the left and right lens. Vertical color blurring can help to cover these problems. Also, remember that we are not yet blurring the luminance, which is where we perceive image detail.

# Undoubtedly a little bit of the wrong eye has leaked over during the Anaglyph encoding process
# Subtract out a bit of the wrong eye. Specified in percentage (0-100)
vidL = Overlay(vidL, vidR, mode="subtract", opacity=(leakCorrectL / 100.0))
vidL = Levels(vidL, 0, 1.0, Int(255 * (1.0 - (leakCorrectL / 100.0))), 0, 255, coring=false)

vidR = Overlay(vidR, vidL, mode="subtract", opacity=(leakCorrectR / 100.0))
vidR = Levels(vidR, 0, 1.0, Int(255 * (1.0 - (leakCorrectR / 100.0))), 0, 255, coring=false)

Often there is leakage between the left and right anaglyph images (and thus the left and right luminance images we’ve calculated here), produced by the mastering or encoding. This bit of code corrects for this leakage by subtracting the opposite eye in small amounts (specified in the parameter file.) Since the subtraction is clamped at zero, we then rescale to full range: 0-255 luminance (8 bits.)

# Horizontally blur if needed - reduces fringing from excessive video peaking.
# Skip if = 1.0 (no blur). Otherwise larger value produce more blur
blurRight != 1.0 ? Eval("""
vidR = BicubicResize(vidR, Int(width(vidOrig) / (blurRight * 2.0)) * 2, height(vidOrig))
vidR = BicubicResize(vidR, width(vidOrig), height(vidOrig))
""") : Eval(""" """)
blurLeft != 1.0 ? Eval("""
vidL = BicubicResize(vidL, Int(width(vidOrig) / (blurLeft * 2.0)) * 2, height(vidOrig))
vidL = BicubicResize(vidL, width(vidOrig), height(vidOrig))
""") : Eval(""" """)

In an attempt to make DVDs look sharper, some anaglyph movies have excessive peaking, a video process to enhance detail. This leaves thin white fringes on the edges of color and luminance boundaries. The boundaries of our red/cyan or green/magenta areas are particularly susceptible. While there are doubtless more sophisticated ways to deal with these fringes, in the interest of processing (and coding) speed, an optional horizontal blur is applied to the offending luminance stream(s.)

# YUV color space for chroma operations
vidR = ConvertToYV12(vidR)
vidL = ConvertToYV12(vidL)
vidColor = ConvertToYV12(vidColor)

Although we’ve tried to keep our processing in RGB space as much as possible, AVISynth likes to paint in color on YV12 color space, and so we convert.

# Use our blurred color video to restore colors to greyscale Right and Left videos
# Skip if we've already got a Mono Right source
isMono != "monoRight" ? Eval("""
vidR = mergechroma(vidR, vidColor)
""") : (""" """)
# Skip if we've already got a Mono Left source
isMono != "monoLeft" ? Eval("""
vidL = mergeChroma(vidL, vidColor)
""") : (""" """)

At long last we apply color to the two luminance video streams, except in the case where we already have a perfectly serviceable stream from the 2D version off the DVD or BluRay.

# Final brightness and contrast tweak
vidR = Tweak(vidR, bright=tweakBrightR, cont=tweakContR, sat=tweakSatR, coring=false)
vidL = Tweak(vidL, bright=tweakBrightL, cont=tweakContL, sat=tweakSatL, coring=false)

Chances are good that we may want to do a bit of brightness, contrast, and color saturation correction on each of the two video streams.

# Maybe we already have a 2D version for one of the eyes
isMono == "monoRight" ? Eval("""
vidR = ConvertToRGB(DirectShowSource(monoName))
""") : Eval("""
vidR = ConvertToRGB(vidR)
""")

isMono == "monoLeft" ? Eval("""
vidL = ConvertToRGB(DirectShowSource(monoName))
""") : Eval("""
vidL = ConvertToRGB(vidL)
""")

If we had that handy 2D version, we avoided processing the left or right stream that corresponds to that 2D view. Now it’s time to insert our 2D video in its proper place.

# Swap if needed
swapAnaglyph == "Yes" ? Eval("""
vidTemp = vidL
vidL = vidR
vidR = vidTemp
""") : (""" """)

The user might want us to swap the left and right video streams. Just doing this to be polite!

# Build the Side by Side (SBS) or Top Bottom (TB) combination of Left and Right video
outputFormat == "SBS_Left_First" ? Eval("""
StackHorizontal(vidL, vidR)
""") : Eval(""" """)

outputFormat == "SBS_Right_First" ? Eval("""
StackHorizontal(vidR, vidL)
""") : Eval(""" """)

outputFormat == "TB_Left_Top" ? Eval("""
StackVertical(vidL, vidR)
""") : Eval(""" """)

outputFormat == "TB_Right_Top" ? Eval("""
StackVertical(vidR, vidL)
""") : Eval(""" """)


Assemble both video streams into a single video, either side-by-side or top-bottom, with the choice of whether left or right comes first (left/top.)

# Optionally resize the output video
outputResize == "Yes" ? Eval("""
BicubicResize(outputWidth, outputHeight)
""") : Eval(""" """)

The user might want to resize the final video smaller or larger. Here’s where it happens. The specified sizes are for the final combined single video stream.

# Dub in the sound
AudioDub(vidsound)

And add the sound back in. If the proper codecs are installed, AVISynth should handle PCM, MP3, AC3, 5.1, 7.1, etc…


# Setup our input files
anaglyphName = "F:/3D Conversions/MovieFolder/YourMovie-GM.avi" # Anaglyph video
PureColName = "F:/3D Conversions/MovieFolder/YourMovie-2D.avi" # Video with color info (either Anaglyph or 2D)
monoName = "F:/3D Conversions/MovieFolder/YourMovie-2D.avi" # Possible 2D video for one eye, if not set to "nothing"
SoundName = "F:/3D Conversions/MovieFolder/YourMovie-2D.avi" # Video with the sound track we want

# Maybe we already have one eye's version in 2D already,
# i.e. the DVD or BR has both 2D and 3D versions
# Set to: monoRight or monoLeft or monoNone
isMono = "monoLeft"

This section tells the conversion process where all the input files are located. Most important is the anaglyph file. Second is the video with our color information. If we’ve got a 2D copy, that’s the best source for color; otherwise you can get reasonable color by using the anaglyph video. If we have a 2D video and we want to use it for either the left or right eye view, the third line is where to specify it. If not, you can put anything in this parameter (string.) The fourth file tells the conversion where to get the audio tracks. Typically this will be either the 2D version or the anaglyph version, but you could use any video file with a sound track.

WARNING: all of these files must be synced to frame accuracy. Sometimes the 2D version and the 3D version are not exactly the same, often being different in the opening credits. You should pre-process the files to be frame accurate before running this conversion!

The next parameter: isMono, tells our conversion whether we already have a 2D version that corresponds to one of the eye’s views. You can set this to monoLeft or monoRight to tell the conversion that you have a 2D copy that should be used as the left or right output. If you don’t have a 2D version or don’t want to use it, set isMono = monoNone.

# Swap eyes: either Yes or No
# Note: it is industry standard to put Red on the left eye for RC glasses
# and Green on the left eye for GM glasses
# It would be unusual to set this parameter to Yes
# since the un-swapped arrangement is either Red=Left or Green=Left
swapAnaglyph = "No"

# Output formatting:
# Choices are:
# SBS_Left_First, SBS_Right_First, TB_Left_Top, TB_Right_Top
#
# Meaning Side-by-Side (SBS) or Top-Bottom (TB)
# And choosing which eye is in which position
# This happens after the optional swap (above)
# and is somewhat redundant, but makes the eye choices clearer.
outputFormat = "SBS_Left_First"

# Resize the output video? Either Yes or No
# If set to No, then the output video is either
# twice the width of the input (for SBS)
# or twice the height of the input (for TB)
outputResize = "No"

# If we are resizing the output, specify the dimensions (Int)
# These dimensions apply to the stacked video size
outputWidth = 500
outputHeight = 200

This section deals with the output file. Although you name the output file in VirtualDubMod (File->Save As…), the layout of that file is determined here. If you want to swap the left and right videos in the output, set swapAnaglyph = “Yes”; otherwise it should be “No”.

Next you you need to tell the conversion whether the two output videos should be arranged side-by-side or stacked vertically. In addition you’ll indicate whether you want the left video first (i.e. on the left of a side-by-side, or top of a vertical stack), or the right video first. The swapAnaglyph parameter reverses the meaning of this order.

If outputResize = “No”, then the width and height of the output video is taken from the input videos (which all must be the same size!) For side-by-side format, the output will be twice as wide as the input, but exactly the same height. For a vertical stack, the output will be exactly as wide as the input, but twice as tall.

If outputResize = “Yes”, then you can specify the output width and height.

Be careful with very large dimensions, especially the width, as some codecs can’t handle really big sizes (>2k.)

# How much to blur the color information (Int or Float)
# This is done by shrinking the color video down in size
# and then resizing it back up to full resolution
# producing a blurred full resolution image
# The two decimate numbers are expressed as percentages
# i.e. a percentage of the full resolution to calculate
# the shrunk size. 100 means no shrink, 10 means 1/10 the
# resolution of the original, etc.
# Anaglyphs are only offset horizontally, so the color blur
# should be strong horizontally, but weak vertically
# For films where the cameras were misaligned vertically
# you will need to make the vertical blur greater.
decimateHoriz = 5.0 # Horizontal shrinkage
decimateVert = 20.0 # Vertical shrinkage - can usually be much bigger than decimateHoriz

As part of the conversion, the videos will be re-colored. This color is extracted from the file you assigned to PureColName. Because the anaglyph will have left and right videos shifted horizontally depending on the depth and strength of the 3D, we can’t be exactly sure where to re-color. Instead we blur the colors so that they approximate the proper location, relying on human’s perceptual weakness for color detail. The blur is achieved by shrinking down the color video and then expanding it back up to full size. There are separate parameters for horizontal and vertical shrink because 3D anaglyphs are displaced mostly in the horizontal direction and therefore more horizontal blurring is needed. The vertical blurring can help to compensate for vertical misalignment of the cameras, lens distortions and other vertical artifacts.

# In case one anaglyph eye has leaked into the other
# We can try to remove that leakage by subtraction
# Expressed as percentage (int or float) (-100 to 100) (0 means none)
leakCorrectR = 10 # Leakage of left into the right eye
leakCorrectL = 0 # Leakage of right into the left eye

# Option to horizontally blur the left and right videos,
# just before the color is restored (before optional LR swap)
# Helps remove some of the fringing that appears in poor DVD encodes
# Set to exactly 1.0 for no processing (faster!!),
# > 1.0 blurs... try 1.5 to 4.0
blurLeft = 1.0
blurRight = 2.0

Here we attempt to correct for leakage of left into right and visa-versa. If you see ghosts of one side appearing in the other side, try subtly adjusting these parameters. Of course, with a great BluRay anaglyph, no correction may be needed at all.

Some anaglyph videos, especially those from DVD or VHS sources will show some fringing around the anaglyph color boundaries after separation by this conversion process. The fringes can be minimized by a slight horizontal blurring.

# Final brightness and contrast adjustments
tweakBrightL = 0 # Left brightness, integer to add to each pixel (pixels are 0-255)
tweakContL = 1.0 # Left contrast adjustment (1.0 means no contrast adjustment)
tweakSatL = 1.0 # Left saturation adjustment (1.0 means no saturation adjustment)

tweakBrightR = -50 # Right brightness, integer to add to each pixel (pixels are 0-255)
tweakContR = 1.35 # Right contrast adjustment
tweakSatR = 1.3 # Right saturation adjustment

Often the anaglyph conversion process will leave the videos looking a bit washed out, or dark, or desaturated. Here is your opportunity to do some basic color correction on either the left or right video (or both!)

# Common code to do the conversion
# Make sure this file path points to
# the file on your system.
import("F:\3D Conversions\AnaExtract.avs")

And finally at the end of the parameter file, the actual conversion code is invoked. Just make sure the path to AnaExtract.AVS is correct for your system configuration. You can put it where ever you want, but you’ll need to change the final line of the parameter file to point to the proper location in your file system.