Hi and sorry if i will say silly things.
I am very ignorant on the subject.
But i read an article saying that to be able to read for instance a car number plate from a satellite the original resolution is not enough.
But applying some sort of digital processing (i saw mentionend the term "fractal", without understanding much) it is possible to improve the actual resolution and in the end read the number 😱
This is extremely amazing
I wonder if anything similar could be possible for instance with a digital MP3 file.
Increasing its resolution with a particular processing methodology.
Would it be possible ?
The interesting thing is that files of small size could sound maybe even better than a CD ?
Image Processing | Geoinformatics @ IITK
Thanks a lot.
Regards, ginetto
I am very ignorant on the subject.
But i read an article saying that to be able to read for instance a car number plate from a satellite the original resolution is not enough.
But applying some sort of digital processing (i saw mentionend the term "fractal", without understanding much) it is possible to improve the actual resolution and in the end read the number 😱
This is extremely amazing
I wonder if anything similar could be possible for instance with a digital MP3 file.
Increasing its resolution with a particular processing methodology.
Would it be possible ?
The interesting thing is that files of small size could sound maybe even better than a CD ?
Image Processing | Geoinformatics @ IITK
Thanks a lot.
Regards, ginetto
Last edited:
Hi,
If a program knows its specifically looking at a numberplate, it
can digout the most likely number from an unrecognisable mess.
However the mess must include enough information, and
the result has no more information than the original file.
The resultant number is a best guess with a certainty factor.
It cannot reconstruct what the numberplate actually looks like.
(Though based on rules it can predict what it should look like.)
With music you don't know what your reconstructing so
you cannot improve it. MP3 has a set of construction
rules that allow the reduced data rate encoding.
You can't get any more information out of the data.
rgds, sreten.
If a program knows its specifically looking at a numberplate, it
can digout the most likely number from an unrecognisable mess.
However the mess must include enough information, and
the result has no more information than the original file.
The resultant number is a best guess with a certainty factor.
It cannot reconstruct what the numberplate actually looks like.
(Though based on rules it can predict what it should look like.)
With music you don't know what your reconstructing so
you cannot improve it. MP3 has a set of construction
rules that allow the reduced data rate encoding.
You can't get any more information out of the data.
rgds, sreten.
Last edited:
Hi and thanks a lot for the very kind and helpful reply.
I was clearly missing something of the process.
Issue closed.
Regards, gino
I was clearly missing something of the process.
Issue closed.
Regards, gino
Numberplates (like many things in real life) contain lots of redundant information. This allows an algorithm to predict what it can't see from what it can see. The snag with this is that you can never be certain that you got it right.
Music contains redundancy too, which is partly why data compression works. The snag is that the more predictable the music, the more boring we find it. Hence data prediction would only work well with music which is not worth listening too.
Music contains redundancy too, which is partly why data compression works. The snag is that the more predictable the music, the more boring we find it. Hence data prediction would only work well with music which is not worth listening too.
Numberplates (like many things in real life) contain lots of redundant information. This allows an algorithm to predict what it can't see from what it can see.
The snag with this is that you can never be certain that you got it right.
Music contains redundancy too, which is partly why data compression works. The snag is that the more predictable the music, the more boring we find it. Hence data prediction would only work well with music which is not worth listening too
Hi and thanks a lot for the very interesting explanation
But another question instead ... given the same compressed MP3 is possible to get different quality of sound using different players SW ?
thanks a lot again, gino
Some of what is going on when increasing the resolution of a digital image is "deconvolution" or "deconvolving filters". This does not necessarily require the application to "know what it is looking for". There are some incredibly powerful deconvolution algorithms in use today.
https://en.wikipedia.org/?title=Deconvolution#Optics_and_other_imaging
Could this be applied to a digital audio bit stream? I don't think so...
https://en.wikipedia.org/?title=Deconvolution#Optics_and_other_imaging
Could this be applied to a digital audio bit stream? I don't think so...
Number plate recognition is made easier because in the UK (and I suppose most other countries) only one font is legal.
This way a lot can be successfully predicted by just a few data points.
If a police cars automated number plate recognition system can't read your plate they will pull you over. If you had plates made in a different font you are technically driving without an MOT and consequently without insurance either. This results in a fine, some points on your licence and when it comes to insurance renewal higher premiums.
This way a lot can be successfully predicted by just a few data points.
If a police cars automated number plate recognition system can't read your plate they will pull you over. If you had plates made in a different font you are technically driving without an MOT and consequently without insurance either. This results in a fine, some points on your licence and when it comes to insurance renewal higher premiums.
Could this be applied to a digital audio bit stream? I don't think so...
The mp3 encoding is not linear or time invariant and what is left over is not uncorrelated noise, all things that would trip up the usual algorithms.
Hi
Why not start with a better codec like Ogg if small file size is needed or Flac
arguably the best of the bunch ?
If you read up on Mastering audio which should be about trying to provide you the listener
with as close as possible to the original sound,you will find good engineers
always insist on the best original source which would not/never include MP3
as it uses lossy audio data compression. A different if not better approach
is used by ECM where the recording is the key to success, rather than
post alteration. https://en.wikipedia.org/wiki/Jan_Erik_Kongshaug .
DBX and Dolby have since the 1970's generally improved the
inability of reel to reel recorders to capture full dynamic range, and
more recently the ability of digital to overload see:
ftp://ftp.dbxpro.com/pub/pdfs/WhitePapers/Type IV.pdf
A Type 1 DBX 150x when used with a hard disc recorder like a yamaha
cdrHD1500 is a reasonably cost effective way of experiencing the
benefits of higher dynamic range, and works exceptionally well with CD
providing dare I say an almost analogue sound to that medium, and
certainly far more acceptable. You need though to spend about $40.00
upgrading its dual and single op amps to LM4562 and LME49710.
Cheers / Chris
Why not start with a better codec like Ogg if small file size is needed or Flac
arguably the best of the bunch ?
If you read up on Mastering audio which should be about trying to provide you the listener
with as close as possible to the original sound,you will find good engineers
always insist on the best original source which would not/never include MP3
as it uses lossy audio data compression. A different if not better approach
is used by ECM where the recording is the key to success, rather than
post alteration. https://en.wikipedia.org/wiki/Jan_Erik_Kongshaug .
DBX and Dolby have since the 1970's generally improved the
inability of reel to reel recorders to capture full dynamic range, and
more recently the ability of digital to overload see:
ftp://ftp.dbxpro.com/pub/pdfs/WhitePapers/Type IV.pdf
A Type 1 DBX 150x when used with a hard disc recorder like a yamaha
cdrHD1500 is a reasonably cost effective way of experiencing the
benefits of higher dynamic range, and works exceptionally well with CD
providing dare I say an almost analogue sound to that medium, and
certainly far more acceptable. You need though to spend about $40.00
upgrading its dual and single op amps to LM4562 and LME49710.
Cheers / Chris
Last edited:
But another question instead ... given the same compressed MP3 is
possible to get different quality of sound using different players SW ?
thanks a lot again, gino
Hi,
AFAIUT the effectiveness of MP3 decoding can and does vary.
As does the encoding you use when ripping files to MP3's.
Its all very complicated and I don't pretend to understand
the minor differences, just that it generally works quite well.
rgds, sreten.
Hi,
AFAIUT the effectiveness of MP3 decoding can and does vary.
As does the encoding you use when ripping files to MP3's.
Its all very complicated and I don't pretend to understand
the minor differences, just that it generally works quite well.
rgds, sreten.
Yes, it is fairly widely accepted that different CODECs (which are employed by different players) will produce different bit streams from the same MP3 file. People often compare LAME decoding to Fraunhofer IIS, for example. They also compare the encoding efficiency and subjective audio quality of the encoding and decoding. By all accounts, there are very real differences. LAME and Fraunhofer are just two of many variations of MP3 CODECs.
The OP should not, however, extrapolate this to lossless formats such as FLAC, APE, etc. With these formats, the bit stream will be identical regardless of the software/CODEC used.
Getting back to MP3; it was primarily designed to be used with VBR and many people make the mistake of configuring their encoding to 320Kbps CBR "for the best quality", when in fact better quality can be achieved with 256Kbps VBR.
This is somewhat moot, however, because better compression algorithms exist (if that is what you are after) and for the best quality a lossless CODEC is the obvious choice.
Last edited:
Hi,
AFAIUT the effectiveness of MP3 decoding can and does vary.
As does the encoding you use when ripping files to MP3's.
Its all very complicated and I don't pretend to understand
the minor differences, just that it generally works quite well.
rgds, sreten
Thanks a lot for the valuable reply.
This is well beyond my limited knowledge but i am very fascinated by digital.
I am of the team analog sounds more musical, natural.
But sometimes some digital units/systems appear that make me think differently.
I heard very very good sound from some digital rigs.
I have very little free time these days but when i will be retired, and hopefully not too retarded, i will look into this more seriously.
Audio needs dedication.
I also think that good digital at reasonable price is quite possible.
However systems based on pc are not completely straightforward IMHO.
Thanks again, gino
Ran into this from Linkwitz' site:
https://xiph.org/~xiphmont/demo/neil-young.html
If he weren't male and ugly, I'd kiss that guy. I notice that he used one of my favorite memes as well.
If he weren't male and ugly, I'd kiss that guy. I notice that he used one of my favorite memes as well.
I get get tired of people tripping over how to spec DNR. Total rms noise to full scale single tone is just so wrong. The Audacity development team lists LP's as 9-10bits or so due to the 65dB performance measured this way.
- Status
- Not open for further replies.
- Home
- Member Areas
- The Lounge
- Improving resolution with digital processing