Merits of the delogo filter

[lordsmurf]

split off from another thread ................


Delogo rarely works well. It tends to just make a blurry blob, not remove the logo.

[AlanHK]

Delogo rarely works well. It tends to just make a blurry blob, not remove the logo.

Delogo works almost perfectly on alpha-blend logos.
I prefer NoLogo (Logotools), which has similar results but is much faster to set up; just requiring one screencap of the logo.

For opaque logos I use Xlogo, (for which I wrote a little guide.). Not so perfect, but still a blur is an improvement on a coloured, distracting logo.

[lordsmurf]

I've never seen a Delogo job that did not leave behind equally annoying blocks, blurs, sparkles, isolated mosquito noise or other artifacts when it was run. I've had people show me their work a number of times, and during the course of the video, your eye was drawn to something weird from time to time. Those filters just don't work too well, aside from short clips.

I prefer a logo over a blob on screen that has it's own life, moving around like a caged amoeba.

Anybody who wants to logo the video shouldn't worry about this. Just my thoughts on it.

All this said, I'm against logos anyway.

[AlanHK]

I've never seen a Delogo job that did not leave behind equally annoying blocks, blurs, sparkles, isolated mosquito noise or other artifacts when it was run.

If the area the filter works on is defined correctly, there is no sparkling, etc.

"Equally annoying"? Maybe to you. For most people a small area of blurring is preferable to a white, or worse, brightly coloured logo. And alpha blend logos are almost perfectly reversible.

Did you even look at the links I gave? Of course not, you already know that I'm wrong.

[lordsmurf]

I looked at the delogo links, but stills are worthless. Show me some 10-minute motion captures at DVD-Video resolutions. I've never seen one that didn't look like a miniature Predator wasn't crouching in the corner of my screen.

[AlanHK]

Show me some 10-minute motion captures at DVD-Video resolutions. I've never seen one that didn't look like a miniature Predator wasn't crouching in the corner of my screen.

If you prefer to see a big bright logo to a pale shadow, that's your choice.

[lordsmurf]

Show me some 10-minute motion captures at DVD-Video resolutions. I've never seen one that didn't look like a miniature Predator wasn't crouching in the corner of my screen.

If you prefer to see a big bright logo to a pale shadow, that's your choice.

I prefer the less distracting object. A static logo vs a moving blob? Give me static.

This comes back to what I told MOvieBuff2 many posts ago. Don't have distractions on screen (colorful, wordy, moving logos), but something simple. THe MB2 logo I made, for example, is a decent non-distracting, non-invasive logo.

But again, I'm against logos, there are better ways to mark up a video. However, if you're going to be bullheaded and slap on a logo, at least do it the correct way.

[AlanHK]

I prefer the less distracting object. A static logo vs a moving blob? Give me static..

"Moving blobs"?
What lousy filter have you been using?

[lordsmurf]

I prefer the less distracting object. A static logo vs a moving blob? Give me static..

"Moving blobs"? What lousy filter have you been using?

Every filter, including the ones you discuss, are lousy filters.

The only way to really remove a logo like that is to go frame by frame and paint it out.

[mats.hogberg]

I took the liberty to split your discussion regarding the removal of logos out to a separate thread.

/Mats

[AlanHK]

The only way to really remove a logo like that is to go frame by frame and paint it out.

No one said "REALLY REMOVE". But 95% is easily done in many cases.
Moving blobs? No, not if done right.

The difference is between an obtrusive logo you can't ignore, and a faint remnant you can detect if you look for it.
You don't have the choice to "really remove". You can't have a pony.

[AlanHK]

.. [posting error]

[lordsmurf]

You still haven't posted an example. Your words mean nothing.

[vhelp]

Although I haven't actually done any 'logo removing' programming projects of my own, I've
always wanted to take a stab at it and add it to the list of things to do one day. I mean,
it would seem its pretty easy to do, I think.

But, working with logo removal is very close to image filtering -- the pixel manipulations are
based on simple math principles, though I don't actually know what math the delogo algos
require.

But I'm guessing that you need to start with two images.. one original (A) and the other
one (B) made the inverse of (A), and then you apply either a Subtraction or Difference to
the two and somehow the logo is somewhat removed. Of course, the algo would intail
the params for the image coordinates location of the logo. Still, I don't know. I'm just
theorizing on this over here at the moment. I don't know.. maybe I'll add it to my filter
template and give it a go on a slow day or something. Anyway.

lordsmurf I know how you feel about logos.. they are the evils of devils I tell ya.
They were put there to annoy us (or those who copy vidoes or whatever) its more like,
payback or something. Specially those large (animated) logos that alert you of a show
coming up next or slated for a certain date. I mean, they just aired a short preview of
it in the prev commercial. Then, half a minute later, they animate it (again -- to annoy
us) and blurt the preview to come. There's no good in all that nonsence!

Regarding the logos.. I think that part of the reason why you see left over gibblets is
because of the noise around pixels within the logos' general area, as this is hard to lock
onto with an algorithm designed to 'reverse' the opque'ness of the logo onto of the video
image.

The (logo'ed) pixels all have to be the same color within the r/g/b color components, and
the image itself has to be as well. If the video source is from an Analog type source, then
you already know how almost impossible it will be to remove, because of the SnP noise
surrounding the whole image, randomly. Then there is the other aspect of this nonsence,
the YUV color space vs. RGB, for when the video is decoded during capturing and/or video
processing. These attributes *add* to the nonsense, compounding it worse.

I would think that if the source was something from a digital medium, like an mpeg from a
satellite or digital cable, you would have a clean non-distorted image, and most de-logo
algorithm tools should be able to remove them. I would assume such. But I could be wrong.

Case some don't know..

From my understanding of video imaging and logo removing, the process should be applied
to the original source -- not resized or filtered, and deff not ivtc'ed either. You have to
start with the original source, as captured, and without hardware or other external filtering..
and that includes, color adjusting as well. For best results, always use the raw source as
the primary for logo removals. And, during the attempt of removal, it should be done inside
the sources original frame rate. That means, if its an interlace or telecine video, it should
be applied during this setup, raw. If you applied an ivtc to a given video and then apply a
logo removal, you loose the videos' frequency consistancy/pattern because the frames
were removed and the pixels no longer match those frequences. This will most likely result
in gibblets and things. Anyway.

Digital (mpeg) vs. Analog video sources are two different beasts. One has a lot of random
noise, otherwise known as Salt n Pepper (SnP) and is beastly to work with in some cases.

Here are a few videos (from the hit tv sereis, That 70's Show) when it was aired by WB.
The source was from my JVC vcr unit, and source was from Analog areal Antenna TV, and
captured by my ADVC-100 dv box. The video is where Eric was waring a red-checker shirt
and he was rainsing his left arm up and then down. I then ran the dv video through vdub
and Ctrl+1 to clipboard and then pasted into ms paint, where I exactly measured it
for a (560x380, 96x52) screen coordinate copy and then pasted into another ms paint and
finally, saved as a bmp file .. for consistancy -- source, as captured back in 11/21/2005.

PARTIAL IMAGES:








** image snipped using pixel coordinates: (560x380, by 96x52) of original images (below)


FULL IMAGES:







** (note, all images are in PNG lossless format)

If you'll notice, in this video sample clip of the logos, there was some 'ghoasting' from the
areal signal. Just another piece of the nonsense of trying to remove the logos.

Oh, and by the way.., the image demonstartes my long-time plauge of Line Noise from that
old ERA.

-vhelp 4466

:P *~*!*~*!~* HaPpY HoLiDaYs *~*!*~*!~*

[AlanHK]

You still haven't posted an example. Your words mean nothing.

I did post links. Are you blind?

If you want me to post megabyte video files, no, I'm not going to spend hours making files and uploading them trying to impress you. I've seen, in many threads here, that you will never change your mind regardless of evidence. When contradicted you just become more doctrinaire and insulting.

[lordsmurf]

Your links are still images. Still images are lies, when it comes to complex video. You miss 99% of the situation. I've seen many stills that look great. Big deal. I have never seen a video with a logo removed with any degree of success. It's just a big ass blob moving around.

[themaster1]

I don't share the same opinion than vhelp i like logos that are well done.
Take the newest abc's logo for example..this is a great logo to me & therefore i don't see the need to remove anything.

[Undead Sega]

has anyone tried MSU Logo Remover???

[AlanHK]

Your links are still images. Still images are lies, when it comes to complex video. You miss 99% of the situation. I've seen many stills that look great. Big deal. I have never seen a video with a logo removed with any degree of success. It's just a big ass blob moving around.

No.

And don't use words like "lies". Lies are deliberate untruths.
I spent a few hours making a how-to for colour logos, copiously illustrated. Anyone can follow that if they have the inclination.

However, the more common alpha-blend logos are even easier to remove and give almost perfect results (using Nologo or Delogo). There are no "blobs" except when removing large colour logos on low-res video. And even in that case, the result, while imperfect, is less annoying than the unfiltered version.

You seem to see everything in black and white. If it isn't perfect, it must be crap.

Processes like this reduce the crap.

[lordsmurf]

Let me put it into perspective.

Somebody wants to set you up with a friend. You're told this person is really cute, and you'd have a great time together. You're skeptical, as well you should be, as few people like a blind date. He says he'd be glad to show you a picture.

He hands you a picture of her chin.

That's basically what you're doing when you give out a still image of a video, and you're trying to show examples of how filters work. You don't get to see much at all, and what you do see is easily an incomplete picture. Most of the time, this incomplete picture is misleading or an outright lie.

Going back to the analogy, it could be a hunchback* with a pretty chin.

I've seen lots of great stills, but every time I get the video of that still, it's crap. Blobs, noise and amoeba.

"If it isn't perfect, it must be crap."
I'm a lesser-of-evils kind of person. Making an annoying blob on screen is worse than just leaving the logo alone. This is just one of those things you can't fix.


* I have nothing against hunchbacks. But let's be real, they probably won't win a beauty contest.

[AlanHK]

He hands you a picture of her chin.

You are implying I chose images to give a false impression. No, I did not. I showed best and worst case images. Below some of the better ones. The others are at the howto page.

Before


After


If you want to enjoy the corporate branding in all its multicoloured glory, fine. Those of us who would prefer to watch a drama with fewer distractions can do so.

[vhelp]

I don't know. I was searching around for some information on how to go about implementing
an algorithm (any) in one of my image analysis tools to see how removing a logo is done. I
still haven't found any code snips or pixel equations to look over, yet. But I haven't given
up just yet. Anyway.

In my search, I came across this article and found it to be pretty exhaustive as a how-to,
using their LOGOAWAY filter plugin for VirtualDub. See the page here:

--> http://www.voidon.republika.pl/virtualdub/ladocs301/logoaway.html
--> LOGOAWAY - A VirtualDub Filter, by Krzysztof Wojdon 2000-2001

Download latest (v3.01) Logoaway release - 9 859 bytes.
and documentation for 3.01 (this HTML page) Logoaway 3.0 documentation - 182 106 bytes.

-vhelp 4468

:P *~*!*~*!~* HaPpY HoLiDaYs *~*!*~*!~*

[lordsmurf]

vhelp, that link brings up a good point. The people who made these anti-logo filters acknowledge that the filters don't work well. And then they were created for a time when VCD and other low-resolution formats were being done at home. The blurring was covered up by even more noise, such as macroblocks and soft image quality.

At high resolutions of DVD and beyond, those filters are incredibly inferior and fail to work well at all.

[vhelp]

You know, I still haven't found any info on algo's and what-not on how to de-logo'ize
images. And so far, google has not been my friend, haha

I'm in the middle of composing some things to post in a follup-up response, but it will take
a little time to put together. It seems to becoming lengthy but necessary. Perhaps
it will help once and for all to realize certain things.. why they are and why we don't see
them when we processing our videos through various (filter) applications.

I have to stop and take a break and make dinner, too. I'll come back a little later on.
So, until then..

-vhelp 4469

:P *~*!*~*!~* HaPpY HoLiDaYs *~*!*~*!~*

[gadgetguy]

I have experimented with logo removal a great deal and for the most part, I agree with lordsmurf about what's left behind being more distracting than the original logo. Having said that, IMO the delogo filter does a really good job of removing an alpha blended logo. But only if you have a good source to start from. I seldom have a source that's good enough to remove the logo without using "repair". Using "repair" does leave blended "blobs" and to my eyes makes it more distracting than the original logo.
However there have been rare occasions where the signal was good, and the logo was transparent enough to not require "repair". On many frames, maybe even most frames, it completely removes all evidence of the logo. Unfortunately, even in the best conditions, it does not completely remove the logo on all frames and whenever the video switches from "clean" frames to "dirty" frames, there appears to be motion where the full logo used to be. In certain scenes it makes no difference because there is enough action occurring in the rest of the scene to render it un-noticeable, but at times it's distracting. I wouldn't describe whats there as a "moving blob", but there is enough distraction there to draw the eye away from the intended center of attention.

http://ericbt.home.comcast.net/SourceSample.m2v
http://ericbt.home.comcast.net/DestSample.m2v

The capture setup for the source for these clips was cable-> vcr tuner -> camcorder passthru -> WinDV -> DVAVI Type1. The first clip was encoded without any filtering. The second I used the DeLogo filter in an AviSynth script then cropped the borders and resized for 16:9 encoding. I chose this particular clip because this area demonstrates where the results are the worst of this entire episode. The remnants of the logo are clearly visible on several of the frames included in this clip. I would not characterize this residue as a "moving blob", but it is there. Whether or not it's noticeable or distracting is in the eye of the beholder.

[vhelp]

Yeah. I'm on no one's side, here. I just look at things a little differently, I guess. Anyway.

I have noted this, time and time again, and I guess, I'm doing, yet again, for the umph'teenth
time but I'll try it in another perspective..

I do think that there is a missing point or aspect to all this de-logo 'ing of images. And it all
has to start with the graphics card features and performance. You'll see what I mean, shortly.

I think that the developers (GOD bless them) were obviously reviewing their images during
regular play inside their software players. These players are all at fault to some degree, in
the aspects how they represent the final output when played inside a certain color space,
though mainly, YUV. But let me try and explain..

First, there are two ways of displaying video images, using:

A) RGB color space
B) YUV color space

Now, if the graphics cards (as most todays generation are desgined) in the usual and same
way for processing the outcome for video in RGB or YUV color space, but the issue becomes
more critical when the video is to be displayed (as it should be) in YUV color space, and usually
with 4:2:0 chroma sampling. So, we are probably talking about YUV 4:2:0 and using the YV12
chroma sampling format, at least, inside such containers.. (ie, XviD; DivX; H264; Flash; WMV;
MOV; MPEG; etc etc) though in reality, this YUV color space is actually converted to RGB in the
end, anyway, weather by your tv set or software player or video editor.. and then the video is
transformed with a slightly higher GAMMA or brightness to give that illusion of YUV color space.

Bare with me as I try to explain a crutial part of all this nonsence and hopefuly, open your
eyes to an understanding of why many are blind to certain key elements in video imaging
and displaying and consiquentially, Review and Analizing and Calculating final output results.

When a player displays an image or video on screen in RGB color space, the eye's mainly do
not see any descirnable problems or artifacts with the video. The RGB color space (thanks
to the GAMA level settings and a few other video attributes not mentined here) the video
as you know it, is mainly dark. Most detail (including the reality of pixelation; DTC; and other
pre-source existing anamilies) are hidden -- not that they are purposely hidden -- but that they
are coensidentailly so. And, over the years, we grew acustom to this phenomina, abliviously,
though not knowingly. So, when we judge/analyize video/images, this hidden fact is a part of
that, the results of the outcome of what we see and consider, quality 'wise, etc. And this
is how its been, since the day we all began our ventures in this hobby.

Now, for me, it was some time in 2001 when I realized this phenomina that I'm still, to this day,
trying to explain as best I can. And, i'm not doing so well at it

I've come into contact with many graphics cards. I've seen how they all produce the same
level of quality in a given image or video, on screen. Most of the cards today are geared
toward Gamming or a combination of compremised features. They are mostly concirned with
the ability to produce Billions of colors now. It is my opinion that this has hinder the true
ability of a graphics card to reproduce video/image in their supossively true nature, or perhaps
they had intended this all along. I don't know. I'm just speculating at this point.

Now, in the very early days of ATI 's development, they put out a unique and special kind of
graphics card. It had the early Theater chip in it.., the first one -- no name, just Theater.
That is the cards chip that I have inside my graphics card. And that card is still in use today
inside my main pc where I process many videos and images for accurate and true image detail
or attributes. The card was produced some time in 2000 or 2001. After that, all the other cards
became modeled but only after an updated chip that was suppose to superseed the first geneartion
of the Theater chip. Now-a-days, these newer generation graphics cards are equipt with the
same 'attribute' features that are responsible for displaying image on a given pc monitor. But
those quality attributes are the same across the board -- graphics card, boards. And this
means that they are not capable of reproducing the image as it should, like the original ATI
Theater chip. They are now based on an permanent template chip, and any upgrades done
from that point, onward. So the features that were once in the orignal template, are long
gone and forgotten. So, you never see an image for what it really is, when its missing some
of the orignal code from the first Theater chip. Weather this was done on purpose or not,
I don't know. Maybe they realize that it would be wiser to deisgn these cards that way, so
that in most case scenarios, the video card will hide or elude the real image for what it is
in reality, in terms of detail and what-not.

So having the knowledge of this history, and when working with your videos, you should realize
by now that you are not seeing the whole truth about your video's image descirnable detail,
especially when reviewing it inside the RGB color space. Now when you review a video inside
a YUV color space, this has the tendency to be slightly brither looking and consiquentially,
you might see a little more of actual image detail and possibly be able to discirn some interesting
nonsense going on in your video. But for the most part, you will not see much more than that,
if any.

Now, the secret (or truth) is this..

My ATI graphics card has a specially unique feature when turned on for YUV color space. I
usually use TMPGenc for this because it also has a unique feature when using the internal
[x] Overlay (for yuv color space) together. It is here where I am revealed the true nature or
aspects or detail of a videos image, frame by frame. And when I want to review a certain
(posted) Video or Image, I use TMPGenc for the partial analisis review.

I can't tell you how many, many Videos and Images I have seen the true nature thereof. Its
been very depressing to say the least. I mostly see many Pixelations; DCT errors; Macro
blocks; Floating or Dancing blocks; Blotchy blocks; Rainbows; Haringbones; you name it. But
I never go further than that to boast or coral anyone's work, when I know that they spent a
lot of time on them.

In my graphics card, and because of the Theater's chip's design features, I am able to see my
video in greater clarity than the rest of you all. When you post sample of Images or Videos,
I can descirn even the slightest of abnormality. Its just like looking at it on tv, but without
going through the laborous process of Authroing a DVD and then finally playing it in a real dvd
player and reviewing on an actual TV set. That's the way my graphics card operates when
inside YUV color space on my main pc. And when I watch a commercial DVD movie, (using
powerdvd) its just like watching it on a TV set. So, I don't need the full use of the TV set
to review a possible video for final archived DVD.

It is because of the above discription that many of you are having difficulty in working out your
videos problem areas, because you can't see those problems to begin with, though your mpeg
compressions, do. And that has many people puzzled and frustrated.. "why is your video
compression greater than mine? ?" ... etc.

Unfortunatelyk, in this odd case, I can not provide actual descirnable proof like you can with
a sample Image or Video. You just have to take my word for it, or not. But I have nothing
to gain by going through all this write-up for you all to read. Plus, I've been saying this for
many years

So now.. the problem with this new knowledge is that Current and New graphics cards will not
(NEVER) incorporate such a unique feature. Its just not practicle or to their advantage to go
and emplement in Newer cards. Its just never going to happen, ever.

..

Now, I took the time to review quicly the two pics posted above. On my main pc, and
inside RGB color space, I could not really see any descirnable differences. I'd 'ov had to
really be looking for something. And, because I would know where to look, it would be
that much more easier. So, I wanted to give a better un-bias review, and so, here is how
I went about honestly reviewing it ...

A few notes about my setup:

I have a KVM switch to control two pc's, hooked up to one keyboard and pc monitor
My main pc has the ATI graphics card
My XP Home computer has a built-in NVidia graphics card which i use as the default

After conveting the two .jpg images to .BMP files, and I rename them to 01.bmp, and 02.bmp,
and then I made two sets, one for my Main PC, and the other for my XP Home computer.

Then, I did the following:

I imported them inside VirtualDub.. on both computers. This made it so that the bitmap files
could be seen as an avi file. So both computers had the same sets of .bmp files.

Then I created a sign-post (psuedo avi file) and frameserved it into TMPGenc.

Then, on both computers, ready and frameserved into TMPGenc, I proceed to turn on it's
Overlay feature, which operates inside YUV color space (though, internally for image, we
all know that it on-the-fly converts it to RGB so that we can see it, because we can't
see the YUV color space image as it is, in that container format)

And, my results came out like this:

XP Home comptuer, the TMPGenc test proved nothing. I could not decirn anything odd or
bad about the de-logo'ed images. They were in effect, flawless, imho. Even in the YUV
color space, the image removal of the logo was as clean and flawless as one could ever hope
for.

My Main PC, however, releaved the opposite. I could see the blotches or ring or something
that resemble what was left behind from the removal of the logo. It was definately there.

Although this demonstration was crude, (two pics) it was very limited and unsatisfactory.
But it did prove a few points.. that most graphics cards of today (and tomorrow) will hide
or elude the true attributes of a given video/image. And weather or not inside YUV color
space or RGB, there is no descirnable'ness to be noticed while reviewing a given video.

The argument about noticing various artifacts after a logo is removed by a given algoithm
is obviously subject to personal scrutney and taste, and based on graphics cards that are
not (or lost its ) capablility to reveal accurate image detail as it is presented these cards
of today and tomorrow's techknowledgy.

..

So you see, there. This is why you can't get accurate results from a video filter that is designed
to do a certain task or process to an image, because:

A) the developer has a tipicle graphics card that eludes the truth, and his/her design is based off
results that were reviewed by such a graphics card and is therefore limited in that scope, and

B) weather or not (A) is true, the user who is running a video through a filter process can not see
the actual results of the videos image because the detail (attributes, good and bad) is not desirnable,
as accurately or completely.

That's how I called it

From the Video Workstation of,
-vhelp 4471

:P *~*!*~*!~* HaPpY HoLiDaYs *~*!*~*!~*

[2Bdecided]

DeLogo works well enough with semi-transparent logos.

http://forum.doom9.org/showthread.php?p=903503

There may be some subtle artefacts left behind sometimes, but there are not-so-subtle artefacts covering most MPEG-2 broadcasts here, and leaving the logo there is an even bigger artefact still!

Cheers,
David.