XDepth® High Dynamic Range Video Compression Technology on the Market
Trellis Management
Trellis Management Co. Ltd. presents the world’s first commercially
available and backward-compatible High Dynamic Range Video compression
technology.
Wednesday, 20th August, 2008 – The ongoing research and development
involving XDepth technologies brought a whole new set of solutions to
the light.
Today Trellis Management demonstrates on its XDepth website, at www.xdepth.com
, the world’s first commercially available High Dynamic Range Video
compression technology.
XDepth HDR-Video is a complete, multi-compatible HDR Video encoding
and real-time playback solution engineered to work with any standard
video codec on the market, either lossy or lossless.
Conversely to other technologies, XDepth HDR-Video isn’t “tone
mapping” dependant and offers an outstanding quality to file-size
ratio.
XDepth HDR-Video files can thus be encoded in any known video format,
from Mpeg-1 up to the latest Mpeg-4 and H.264, always resulting in a
backward-compatible AVI file that can be normally played by any
standard Media Player.
The peculiar efficiency of the XDepth HDR-Video encoding technology
allows for complete HDR luma and chroma recovery at a fraction of the
computations required by other solutions and HDR Video AVI files to
always weigh very similarly to their standard 24bit/pixel counterpart.
The XDepth HDR-Video Player is built over the latest technologies and
allows the execution of custom filters over the whole video stream in
real-time.
High Dynamic Range Video compression technologies find their best
applications in advanced surveillance systems, military applications,
high-contrast video capturing for scientific analysis or entertainment
purposes, HDR Display devices, 3D rendering and video editing within
the multimedia production industry.
The availability of XDepth HDR-Video technologies will drastically
reduce the need for expensive high-speed data- transfer interfaces,
enormous amounts of storage devices, solid memories and custom video
codecs, usually required when capturing, compressing and storing video
at non-standard high bit-depths.
Interested parties can contact XDepth personnel and managers for more
information through our contact page at: www.xdepth.com/contact.php .
About Trellis Management:
Trellis Management is a privately held company focused on technology
IP's licensing, acquisition and sale. Trellis Management is not
related nor affiliated with the Joint Photographic Experts Group
(JPEG) or the Moving Picture Experts Group (MPEG).
Industrial One
> XDepth HDR-Video files can thus be encoded in any known video format,
> from Mpeg-1 up to the latest Mpeg-4 and H.264, always resulting in a
> backward-compatible AVI file that can be normally played by any
> standard Media Player.
Whoa whoa, wait, what are you saying? HDR is a revolutionary codec or
CONTAINER? H.264 technically can't go into AVI without extensive hacks
and some features are completely impossible to support. So... if
XDepth HDR employs a revolutionary algorithm superior to H264 then I
don't think an outdated container like AVI would support it.
How much improvement can we expect in bitrate/speed over H.264? What
are the key components of XDepth HDR that makes it so superior?
Trellis Management
Hi,
let me explain this better than how it's been written in our press
release.
XDepth HDR-Video allows "High Dynamic Range" video sequences to be
compressed using any codec into a regular and compatibile AVI.
Now, what's "special" about this is the fact that you normally do not
compress HDR video sequences, at all.
HDR is a special kind of imaging - very much used in VFX, 3D animation
studios and in movies production generally - that allows each pixel to
represent extreme contrast (or, "real lighting" values).
You can find a few examples and external links on our website, but
let's make a "real life" example:
Let's say that you have a picture of a landscape with clouds, sun and
flowers etc. in standard "Low Dynamic Range" (LDR) imaging, the
"white" of the sun is R: 255,G: 255,B: 255 (24bit/pixel) and the white
of a flower in the same picture is still "white", R: 255,G: 255,B:
255.
In High Dynamic Range there is no "white" in this sense... there are
floating point values (pixels) from 0 up to 2^32 for each color
channel (if there is no sign bit).
This means that the "landscape" picture taken and saved in HDR can
represent the flower with R: 1.0, G: 1.0, B: 1.0 while the sun will be
more something like R: 200000, G: 195000, B: 190000 ... (96bit/pixel,
floating point)... maintaining a "digital" contrast between the high-
brightness "white" of the sun and the low-brightness "white" of the
flower - as it happens in reality.
So, given this special type of "still images", video compression has
consequently been quite difficult to achieve till now. The only other
technology that achieves similar results is Dolby's HDR, but it's not
on the market yet.
Clearly, you won't find any video codec around that can do HDR, at the
moment - if not our technology.
There are several advantages in representing images and video using
HDR. So have a look around the internet, and if you got more questions
I'll try and answer.
You can also use our forums...there're some talented people that know
quite a lot about HDR.
Industrial One
Umm, if the flower is truly that bright to be a perfect white (FFFFFF)
then that's just how it is. The sun in reality would probably be a way
brighter white but if a 24 bit color depth is used then the limit is
255 and so it will be in the picture. If you use 96 bits then it'll be
way harder for a flower to reach a maximum value and the sun's pixels
will most likely be higher. I don't see your point. The human visual
perception is very limited and I don't think a 96 bit depth will have
any noticeable higher quality.
What does any of this have to do with creating the same quality videos
at lower bitrates?
Trellis Management
> > Let's say that you have a picture of a landscape with clouds, sun and
> > flowers etc. in standard "Low Dynamic Range" (LDR) imaging, the
> > "white" of the sun is R: 255,G: 255,B: 255 (24bit/pixel) and the white
> > of a flower in the same picture is still "white", R: 255,G: 255,B:
> > 255.
> > In High Dynamic Range there is no "white" in this sense... there are
> > floating point values (pixels) from 0 up to 2^32 for each color
> > channel (if there is no sign bit).
> > This means that the "landscape" picture taken and saved in HDR can
> > represent the flower with R: 1.0, G: 1.0, B: 1.0 while the sun will be
> > more something like R: 200000, G: 195000, B: 190000 ... (96bit/pixel,
> > floating point)... maintaining a "digital" contrast between the high-
> > brightness "white" of the sun and the low-brightness "white" of the
> > flower - as it happens in reality.
>
> Umm, if the flower is truly that bright to be a perfect white (FFFFFF)
> then that's just how it is. The sun in reality would probably be a way
> brighter white but if a 24 bit color depth is used then the limit is
> 255 and so it will be in the picture. If you use 96 bits then it'll be
> way harder for a flower to reach a maximum value and the sun's pixels
> will most likely be higher. I don't see your point.
We don't need the flower to reach the values of the sun...we need it
to stay how it is.
In common images, the sun *should* be higher, but it's not because you
only got one "white" point, that is 255,255,255.
HDR keeps the sun brighter even though the flower stays
white...imagine going "over" the white you can see in the actual
picture.
In the LDR version of the "landscape" you can't make a difference
between the two "brightnesses" anymore - while you should (and you
can, using HDR).
This whole HDR technology allows, for instance, to re-expose a digital
picture after it's been shot. You can't do that with LDR images.
It also allows to apply the so called "tone mapping operators", which
are algortithms that attempt to "shrink" the whole range in a pleasant
way, having everything perfectly exposed. You can sort of "see through
glares", since you can take down exposure at your will, and see what's
"behind" the glare, or enhance extremely dark details that would
otherwise be lost.
>The human visual
> perception is very limited and I don't think a 96 bit depth will have
> any noticeable higher quality.
It's not really a matter of "quality". It's a matter of keeping the
lighting values as they are.
Take a look at this picture for instance:
http://en.wikipedia.org/wiki/Image:Dupont-HDR.jpg
You wouldn't be able to shoot that photo as shown without HDR. Your
camera, which saves jpeg or RAW, cannot keep the high-bright sky and
the fountain in the foreground, way darker in comparison, both well
exposed. In this case HDR is obtained by combining several photographs
taken at different exposures. The HDR picture in the link is then
"tone mapped" to 8bit/channel by an algorithm of choice, in order to
be displayed.
Here are a few references for HDR in general:
http://en.wikipedia.org/wiki/High_dynamic_range_imaging
http://debevec.org/Research/HDR/
http://www.anyhere.com/gward/hdrenc/hdr_encodings.html
The easiest way you can experiment with this is checking out our image
gallery at www.xdepth.com/gallery.php using Internet Explorer.
You will be able to change the exposure of the picture in realtime
within IE.
Be sure to let IE execute the ActiveX control - don't worry, there's
no malware of sorts.
> What does any of this have to do with creating the same quality videos
> at lower bitrates?
Not the same "quality" videos, but HDR videos that are originally a
sequence of very heavy floating point images.
Take a 640x480 HDR image sequence of 840 frames (35 seconds at 24
fps).
Not compressed would be 96bit/pixel * 640 * 480 * 840 = ~3.1 GBytes.
Our video brings that sequence to a ~18.5 MBytes avi file, using Mpeg4
as the base codec.
Considering you won't normally compress that type of images, that's
probably quite a good ratio.
Industrial One
a compression codec like I thought, my bad. However, I still don't see
the point of the 96 bit depth. That picture you linked is an already-
compressed 24-bit JPEG image, so I guess this means we can safely
transcode since the only purpose of your technology is to capture more
visually-accurate *source* photos?
Trellis Management
Sorry, misunderstanding again. I described a "capturing" technology in
order to introduce you to "HDR" in general.
We don't "own" HDR as a technology, we own technologies that compress
HDR, both images and video.
As for converting HDR to Jpeg, yes, it is possible to transcode HDR to
24bit images and that operation is called "tone mapping", which
involves a lot more than simply cutting precision on the numbers.
http://en.wikipedia.org/wiki/Tone_mapping
It's an active research topic, together with HDR in general.
Harold Aptroot
news:ed8edb1f-f65a-4448...@25g2000hsx.googlegroups.com...
So why not just do that before compression and get an even smaller file and
even faster playback?
(or am I missing some fundamental point here?)
Eventually these pictures have to be forced into 24bit colour depth anyway
because of the screen..
Trellis Management
> "Trellis Management" <trellism...@gmail.com> wrote in message
If you "tone map" and discard the original HDR it's like throwing away
your "digital negative". You lose the ability to re-process the
imagery using HDR data.
Moreover, while not yet on the market, there will be HDR displays that
will take advantage of the high contrast represented in such images.
That means you will want to keep HDR data up to the final output, in
order to display the content properly on such devices.
Industrial One
> "Trellis Management" <trellism...@gmail.com> wrote in message
Seconded.
George Johnson
news:2c9d67fe-fc19-4365...@i20g2000prf.googlegroups.com...
Let me point you in a direction.
Told you folks here that taking video to holographic (3D object modeling
+ 4D time compression) would save a bundle more space than just basic dumb
2D + time compression. The other nifty bonus is that grayscale movies can
be colorized (or color-improved for crappy 1970's film) using this
computational tech. In theory, you could also take an episode of the
television series "Green Acres" (or any show), build a nearly duplicate set
on computer or just use similar props and this tech should enable the video
editors to vastly improve the look of the old show for almost no cost in
time. Also note that this is EARLY TECH, this stuff will just get better in
another decade. You haven't seen the barest of the interesting stuff yet.
http://gizmodo.com/5038183/photographs-enhance-video-in-absolutely-unbelievable-ways
Photographs Enhance Video in Absolutely Unbelievable Ways
Before Gizmodo, I worked in the bowels of the broadcast industry for a
number of years. I was either shooting video or cutting video every day, all
day. And while Final Cut Pro and Adobe After Effects were both tools I used
with some proficiency on a daily basis, I've never seen a post production
demo as incredible as this clip from the University of Washington.
Essentially, you shoot some crappy, low-rez video of a still scene. You then
reshoot the same scene with a digital camera (with higher resolution).
Software can automagically combine these images to upconvert the video AND
fix problems in the image- all while compensating for 3D space. Make sense?
The remarkable demo will clarify things a bit:
(Video viewable at Gizmodo webpage)
What's especially notable is that the software can fill in the nasty bits of
the scene despite the videographer/photographer rotating their view (you see
this as they shoot around the tree) and despite any lens differences (the
software can compensate for different lens sizes/distortions).
Also, note that many details from the source video are retained (the glass
reflections in the statue shot may be the best example), which means that
the photograph's information isn't the only information we see in the
composite image.
I'm not quite convinced that the entire process is quite as automatic as the
students would make it, but the technology is extremely promising all the
same. And at this point, it should only be a matter of time before we see
the idea work its way into our favorite post production products
http://grail.cs.washington.edu/projects/videoenhancement/videoEnhancement.htm
---------------------
Content Aware Image Resizing
http://www.youtube.com/watch?v=qadw0BRKeMk
(Lossy data compression applications anyone?)
Want to play with this idea online?
Now you can play too online (tried the free software download, but it's not
available or the free server is swamped). FOR REMOVING OR PRESERVING DETAILS
YOU HAVE TO SLIDE THE BIG SLIDERS ON THE SIDES OF THE PLAY AREA TO HAVE YOUR
ALTERATIONS STICK.
Not as hot in removing big image details on small scenes. Had an image of a
beardy fellow riding an elephant, tried removing him while preserving some
line details in the structure behind him, the results were worse than a
cheap sloppy photoedit. I figure the free online version is lacking the
"generate detail" code of the refined versions. On the other hand, it works
pretty damn well despite that.
------------------------------
http://markstechnologynews.blogspot.com/2008/08/video-microsoft-makes-video-vandalism.html
Thursday, 14 August 2008
Video: Microsoft Makes Video Vandalism As Easy As Touching Up A Photo
What's the matter? Photoshopping images of your friends and foes not as fun
as it used to be? Well, Microsoft has come to the rescue with "Unwrap
Mosaics", a new piece of software that makes real-time, in-motion edits to
video faster than you can say handlebar mo.
The software uses masks to add or undo any edits you perform and comes
loaded with editing functions such as cutting/pasting/copying, resizing,
repainting, deforming and effects layers.
In the video below, the software is demonstrated on a head shot of some guy
chattering away. The images that make up the video are converted
automatically to an "unwrap mosaic" and any changes made to the mosaic are
re-composited into the original video sequence.
The video demonstrates the real-time, in-motion application of virtual
make-up including rouge on the cheeks, kohl on the eyes, big bushy eyebrows
and, why not, a big biker mo. Don't know if it's really suited for
professional production just yet, but it sure looks like a winner for a
laugh on amateur vids.
(Displayed video at the website)
Sources:
Microsoft Unwrap Mosaics
Microsoft Unwrap Mosaics: A new representation (Longer video - 6 min 27
secs)
http://www.youtube.com/watch?v=mNLx9pclMKU
http://research.microsoft.com/unwrap/
George Johnson
news:ed8edb1f-f65a-4448...@25g2000hsx.googlegroups.com...
I applaud your work and have to caution you that the "Industrial One" is
generally a jackass (backtrack on GOOGLE groups to enjoy that person's
"colorful" language and an obsession with being a rude jackass in this
group).
http://groups.google.com/groups/search?q=author%3Aindustrial_one%40hotmail.com&start=0&scoring=d&hl=en
This is another interesting bit of research too. It lacks certain
pattern matching routines and vector-tracking recognition, but is brilliant
to say the least. A bit of recognition of 3D layers would help the Seam
Carving tech a bit more too.
Improved Seam Carving for Video Retargeting
http://www.youtube.com/watch?v=AJtE8afwJEg
------------------
Here is another genius team I am watching with glee as they will be
capturing "Smarter Video" versus throwing more brute force computation at
the task.
Programmable Aperture Photography: Multiplexed Light Field Acquisition
http://www.youtube.com/watch?v=8ENfPYpkHp4
http://mpac.ee.ntu.edu.tw/~chiakai/pap
Programmable Aperture Photography: Multiplexed Light Field Acquisition
(SIGGRAPH 2008)
Chia-Kai Liang, Tai-Hsu Lin, Bing-Yi Wong, Chi Liu, and Homer Chen
National Taiwan University
--Abstract--
We present a system including a novel component called programmable aperture
and two associated post-processing algorithms for high-quality light field
acquisition. The shape of the programmable aperture can be adjusted and used
to capture light field at full sensor resolution through multiple exposures
without any additional optics and without moving the camera. High
acquisition efficiency is achieved by employing an optimal multiplexing
scheme, and quality data is obtained by using the two postprocessing
algorithms designed for self calibration of photometric distortion and for
multi-view depth estimation. View-dependent depth maps thus generated help
boost the angular resolution of light field. Various post-exposure
photographic effects are given to demonstrate the effectiveness of the
system and the quality of the captured light field.
--------------------
Making photography SMARTER rather than requiring more image-capturing
elements.
http://www.umiacs.umd.edu/~aagrawal/index.html
I got interested in these fellow's brilliant work with this webpage.
Depth Edges in Real-Time using Multi-flash Camera
Amit Agrawal, Vlad Branzoi, Rama Chellappa, Rogerio Feris, Ramesh Raskar,
Karhan Tan and Matthew Turk
Real time demo at CVPR 2005
Multi-flash (Non-photorealistic) Camera
http://www.merl.com/people/raskar/NprCamera/
Unfortunately, the webpage is not online at the moment.
http://www.google.com/search?q=depth+edges+multi-flash+camera
An example is available to view though
Multi Flash Camera: Non-photorealistic camera, depth edge de
http://www.youtube.com/watch?v=9dLabbCHhZA
===========
http://www.digitalcamerainfo.com/content/Mitsubishi-Electric-Develops-Deblurring-Flutter-Shutter-Camera.htm
Mitsubishi Electric Develops Deblurring Flutter Shutter Camera
August 30, 2006 - Following this month's 33rd Annual Siggraph Conference in
Boston, MA, a research team at Mitsubishi Electric is catching the attention
of camera manufacturers for their photo motion deblurring technology, called
a flutter shutter camera.
The flutter shutter camera is a modified camera that can capture moving
objects at an exposure time of over 50 milliseconds, like high speed motion
cameras. Using a coded exposure sequence, the new flutter shutter camera
could recover text from a speeding car and sharpen images, according to the
researchers.
Introduced in early August, three Mitsubishi Electric researchers presented
the abstract, "Coded Exposure Photography: Motion Deblurring using Fluttered
Shutter" at the largest computer and graphics conference, Siggraph. After
one year of research development, Mitsubishi Electric Research Lab (MERL)
senior researcher Ramesh Raskar, MERL visiting researcher Amit Agrawal, and
Northwestern University computer science assistant professor Jack Tumblin
launched the new prototype with the goal of deblurring photos.
The prototype is made with an 8 megapixel Canon PowerShot Pro1, although it
could be applied to any camera. Instead of leaving the shutter open during
one exposure duration, the camera's attached lens filter flutters the
shutter multiple times during a single exposure, based on a carefully chosen
binary sequence.
Raskar, who celebrates his sixth year at Mitsubishi this month, woke up one
day with the idea, according to the researcher. Raskar said the fluttered
shutter method was "so simple" and wondered if it could work.
"We have UV filters. We have polarizing filters. What about time filters?"
said senior research scientist Ramesh Raskar at Mitsubishi Electric. Just a
few weeks ago, Raskar and the Mitsubishi team saw the dream actualized with
its official introduction. This time filter is made up of an external
ferro-electric shutter that flutters based on a rapid binary sequence.
Traditional cameras typically have a single shuttered exposure in which
moving subjects result in blurry images. On standard cameras, object motion
can be described by convolution of a sharp image with a temporal box filter,
according to researchers, thus destroying high frequency spatial details in
the image. The coded exposure camera, on the other hand, uses a method
called deconvolution, in which the convolution filter changes to a broadband
filter and preserves spatial frequencies and image detail.
A post-capture linear system algorithm is then applied to recover image
sharpness. The algorithm is "fundamentally different from other deblurring
algorithms," stated authors of the abstract. The algorithm, Ax=b (a Matlab
Code), is the simplest deconvolution algorithm possible, according to
researchers. Unlike Photoshop or other deblurring methods, this
deconvolution does not result in ringing or deconvolution artifacts,
halo-like distortions on the image.
http://www.umiacs.umd.edu/~aagrawal/sig06/sig06.html
George Johnson
news:ed8edb1f-f65a-4448...@25g2000hsx.googlegroups.com...
Another bit of research having great applications in visual data
compression at low memory costs are "Diffusion Curves".
Diffusion Curves
http://www.youtube.com/watch?v=lEVe7vU5WiU
Added: April 21, 2008
Video of the Diffusion Curves, from SIGGRAPH 2008.
The video presents the "Diffusion Curve" primitives for the creation of soft
gradients and blur in vector graphics, along with an image analysis method
to automatically extract Diffusion Curves from photographs.
Authors: Alexandrina Orzan, Adrien Bousseau, Holger Winnemöller, Pascal
Barla, Joėlle Thollot, David Salesin
Trellis Management
> "Trellis Management" <trellism...@gmail.com> wrote in message
>
> news:ed8edb1f-f65a-4448...@25g2000hsx.googlegroups.com...
>
>
>
> > On 21 Ago, 01:11, Industrial One <industrial_...@hotmail.com> wrote:
> >> That's different. You're announcing a new photocapture technology, not
> >> a compression codec like I thought, my bad. However, I still don't see
> >> the point of the 96 bit depth. That picture you linked is an already-
> >> compressed 24-bit JPEG image, so I guess this means we can safely
> >> transcode since the only purpose of your technology is to capture more
> >> visually-accurate *source* photos?
>
> > Sorry, misunderstanding again. I described a "capturing" technology in
> > order to introduce you to "HDR" in general.
> > We don't "own" HDR as a technology, we own technologies that compress
> > HDR, both images and video.
>
> > As for converting HDR to Jpeg, yes, it is possible to transcode HDR to
> > 24bit images and that operation is called "tone mapping", which
> > involves a lot more than simply cutting precision on the numbers.
>
> >http://en.wikipedia.org/wiki/Tone_mapping
>
> > It's an active research topic, together with HDR in general.
>
> Another bit of research having great applications in visual data
> compression at low memory costs are "Diffusion Curves".
>
> Added: April 21, 2008
> Video of the Diffusion Curves, from SIGGRAPH 2008.
> The video presents the "Diffusion Curve" primitives for the creation of soft
> gradients and blur in vector graphics, along with an image analysis method
> to automatically extract Diffusion Curves from photographs.
> Authors: Alexandrina Orzan, Adrien Bousseau, Holger Winnemöller, Pascal
Hi George, and thank you for your interest and for the interesting
readings you posted.
We knew some of the material you presented, while some is new to us,
so we will take a careful look at it.
The progress on seam carving looks very nice indeed.
Talking about hardware, there certainly are infinite ways to reach a
given goal, and we are looking into developing our own way for HDR.
Jim Leonard
> Umm, if the flower is truly that bright to be a perfect white (FFFFFF)
The point of HDR is that you are no longer limited to 8 bits per
component (ie. r+g+b=24 bits of information). HDR video uses 10- or
12-bits per component for greater flexibility in post-production. I
used 10-bit on my last DVD, for example.
Industrial One
> "Industrial One" <industrial_...@hotmail.com> wrote in message
>
> news:2c9d67fe-fc19-4365...@i20g2000prf.googlegroups.com...
>
> > On Aug 20, 9:05 am, Trellis Management <trellism...@gmail.com> wrote:
> >> XDepth HDR-Video files can thus be encoded in any known video format,
> >> from Mpeg-1 up to the latest Mpeg-4 and H.264, always resulting in a
> >> backward-compatible AVI file that can be normally played by any
> >> standard Media Player.
>
> > Whoa whoa, wait, what are you saying? HDR is a revolutionary codec or
> > CONTAINER? H.264 technically can't go into AVI without extensive hacks
> > and some features are completely impossible to support. So... if
> > XDepth HDR employs a revolutionary algorithm superior to H264 then I
> > don't think an outdated container like AVI would support it.
>
> > How much improvement can we expect in bitrate/speed over H.264? What
> > are the key components of XDepth HDR that makes it so superior?
>
> Let me point you in a direction.
> Told you folks here that taking video to holographic (3D object modeling
> + 4D time compression) would save a bundle more space than just basic dumb
> 2D + time compression. The other nifty bonus is that grayscale movies can
> be colorized (or color-improved for crappy 1970's film) using this
> computational tech. In theory, you could also take an episode of the
> television series "Green Acres" (or any show), build a nearly duplicate set
> on computer or just use similar props and this tech should enable the video
> editors to vastly improve the look of the old show for almost no cost in
> time. Also note that this is EARLY TECH, this stuff will just get better in
> another decade. You haven't seen the barest of the interesting stuff yet.
Yeah, object-based codecs. I've been hearing about that shit for a
while and gave up on the idea by now when the 4th software company
failed to prove itself -- not to mention how you're pulling shit outta
your ass right now. What does colorization have to do with 3D object
modeling? You would have to seperate a frame into regions where you
add color... whether the 3rd plane can be modeled or not doesn't
change the fact that a program cannot know what color to add if it can
at least automatically model the regions correctly.
> http://gizmodo.com/5038183/photographs-enhance-video-in-absolutely-un...
> Photographs Enhance Video in Absolutely Unbelievable Ways
> Before Gizmodo, I worked in the bowels of the broadcast industry for a
> number of years. I was either shooting video or cutting video every day, all
> day. And while Final Cut Pro and Adobe After Effects were both tools I used
> with some proficiency on a daily basis, I've never seen a post production
> demo as incredible as this clip from the University of Washington.
>
> Essentially, you shoot some crappy, low-rez video of a still scene. You then
> reshoot the same scene with a digital camera (with higher resolution).
> Software can automagically combine these images to upconvert the video AND
> fix problems in the image- all while compensating for 3D space. Make sense?
> The remarkable demo will clarify things a bit:
Re-shoot exactly the same scene? How...? Photo or video? And what's
the purpose of all this, to purify a scuzzy shit-smeared movie, right?
And is this a one-time comparison you feed to the program so it knows
how to clean a different scene with the same defects? Not useful if I
only got one copy and have no access to the other side of the planet
to film the scene for a clean version.
> (Video viewable at Gizmodo webpage)
>
> What's especially notable is that the software can fill in the nasty bits of
> the scene despite the videographer/photographer rotating their view (you see
> this as they shoot around the tree) and despite any lens differences (the
> software can compensate for different lens sizes/distortions).
>
> Also, note that many details from the source video are retained (the glass
> reflections in the statue shot may be the best example), which means that
> the photograph's information isn't the only information we see in the
> composite image.
>
> I'm not quite convinced that the entire process is quite as automatic as the
> students would make it, but the technology is extremely promising all the
> same. And at this point, it should only be a matter of time before we see
> the idea work its way into our favorite post production products
>
> http://grail.cs.washington.edu/projects/videoenhancement/videoEnhance...
>
> ---------------------
>
> Content Aware Image Resizing
>
> http://www.youtube.com/watch?v=qadw0BRKeMk
>
> (Lossy data compression applications anyone?)
>
> Want to play with this idea online?
>
> http://rsizr.com/
>
> Now you can play too online (tried the free software download, but it's not
> available or the free server is swamped). FOR REMOVING OR PRESERVING DETAILS
> YOU HAVE TO SLIDE THE BIG SLIDERS ON THE SIDES OF THE PLAY AREA TO HAVE YOUR
> ALTERATIONS STICK.
>
> Not as hot in removing big image details on small scenes. Had an image of a
> beardy fellow riding an elephant, tried removing him while preserving some
> line details in the structure behind him, the results were worse than a
> cheap sloppy photoedit. I figure the free online version is lacking the
> "generate detail" code of the refined versions. On the other hand, it works
> pretty damn well despite that.
>
> ------------------------------
>
> http://markstechnologynews.blogspot.com/2008/08/video-microsoft-makes...
Liquid retargeting is a start for intelligent engineering but ain't
practical yet. It makes the image look way out of proportion and does
no better than already-existing Lanczos resize algorithms.
Trellis Management
Hi Jim,
HDR in its uncompressed form is a 96bit/pixel floating point image
format. (some common uncompressed formats are Tiff Float, PFM, FLX,
FIM)
A quasi-lossless format is Radiance .hdr, encoded as a 32bit/pixel
image containing 24bit mantissa and 8bit exponent for each pixel (IEEE
float).
The higher bit-depths in image and video processing are mainly used to
reduce the good old banding artifacts when blending multiple sources
to the final output - or execute complex calculations as in a 3D
graphic card.
The HDR concept basically resides in what is "captured" or generated
into the image. If you simply look for "gradient" accuracy, then you
don't necessarily need HDR. If you look at smooth gradients between
hugely "spaced" values, then you actually need to represent the image
as HDR.
As an example, let's say you have (in integer form) as input some
pixel values, for a single component, of 0; 1.000; 100.000; 1.000.000,
and smooth gradients in-between.
The point of HDR resides in having such extreme values (captured or
generated) as input in the first place and the need to store those
without losing the original "spacing" (contrast). Using enough bits in
order to shade the pixel values is a logical consequence and need, but
the higher bit depth alone doesn't necessarily make it an HDR image/
video.
I hope this gave some more "depth" to the argument. :)