Anti noice algorithm

[cyborg]

Hey!
A few days ago, i thought about a way to reduce noice in animations or movies, but its a kind of hard thing to do automaticly...
Yesterday i read that NASA uses this kind of thing to look for stars...


This is how this could be done:

If an object moves from one side of the camera to the other, the noice in the pixels of the object would change because of the movement.

If the program can run the frames and check for certain patterns in mixes of the frames, like lines and for colors thats almost the same on some spots, it could make a final frame where the object is reduced from noice!


Does anyone have an idea how to make this program?
It could be really good....

[anjulpa]

most compression algoriths for video (and for images) already use such algorithms. If you want to develop such an algorithm, develop highly qualifiable mathematics, create the algorithm, look if it isn't patented, check efficiency, patent it, license it, make loads of money.

PS. if you make much money, be sure to make me a friend

[NoteMe]

ANd try to learn C++ or something at the same time...

[cyborg]

hehe! sound like its going to be too hard...

[cyborg]

btw. im learning C++ in school right now

[cyborg]

i also thought about doing a reverse cencur algorithm..
you know sometimes when someone on tv doesnt want to show his fase, they add the pixellerate effect that takes some of the pixels and enlarge them....when he moves a bit, those pixels change...by knowing how large the big "pixels" are, and how much he moved, you can puzzle together those pixels in the normal size and get the face (the more he move, the more pixels he reveals)

[NoteMe]

Hehehe..good luck...

[jim mcnamara]

FWIW - look into the fast Fourier transnform fft -

Numerical Recipes in Basic

Numerical Recipes in C

Books that both have it.

[Zaei]

Originally posted by cyborg
i also thought about doing a reverse cencur algorithm..
you know sometimes when someone on tv doesnt want to show his fase, they add the pixellerate effect that takes some of the pixels and enlarge them....when he moves a bit, those pixels change...by knowing how large the big "pixels" are, and how much he moved, you can puzzle together those pixels in the normal size and get the face (the more he move, the more pixels he reveals)

Wouldnt work... you wouldnt have any data to fill the rest of the pixels in with =P.

Z.

[cyborg]

yes....when the face moves just a little bit the big squares are not moved, so it gets filled with some pixels thats beside the first one...so i just have to plot those pixels beside each other in original size

[NoteMe]

I can see your idea...but how would you know how much the person moved????

[cyborg]

maybe....it depends if the censoring doesnt covers the whole face...else i dont think it would be possible...

[NoteMe]

Yes, but in a "movie" clip that is about 10-15 sec long. It will wary in grreat scale if the none of the face is showing. Or if some parts of it is showimg. And I can't see a way to determin how much that face moved down to pixel performance...

[cyborg]

the face cant move less than a pixel width, so if i can see some of the face...its possible...
or if the whole body moves almost the same as the face, i could do it too...

[NoteMe]

But from picture to picture in the movie clip, he also might move his head arond, and not only to left/right. That would make the lightning diffrent, and complicate the whole thing...

[cyborg]

yep..very much...but im not looking for a perfect reconstructing of the face/object, just an aproximation of what it looks like...

[NoteMe]

And I'm looking forward to see if you cn manage it...are you going for a try????

[cyborg]

maybe...im very busy with the band right now, so it has to be in some weeks or so...

[NoteMe]

If you do, post about it...I'm looking forward to see where this ends....

[cyborg]

sure i will!