|
|
| David.Bucci |
| Posted: Aug 19 2002, 06:13 PM |
 |
|
Unregistered
|
Does anyone know of any free image stabilization filters? The $50 one at, uh, digispeak? hang on, new-mozilla-tab-man is about to save the day .... DigiStudio's Digital Video Stabilizer seems like a great product, for my one-time use $50 is more than i wish to pay.
tia |
|
| bobsobol |
| Posted: Sep 17 2002, 12:12 AM |
|
|
Unregistered
|
A temporal cleaner can be very effective at low settings on shots with low levels of jitter. (½ to 1 scanline fe)
Greater than that, I would suggest getting a camcorder with jitter correction built in, or (for pre-recorded material) pay the 50 bucks, do it by hand (Paint Shop Pro / Animation Shop & an old old copy of Animation Studio work well for me) or dare I say...find a warez. 
The "do it by hand" method dose mean manually modifying each frame, so you don't want to edit more than a few seconds. Place one frame in the senter of a canvas about twice as wide and twice as high. (this is slightley excessive but safe) keep this frame for referance against each sucessive frame.
Place each sucessive frame in a new layer over the top of it, set this layer to 50% transparency and match key features you want to "stop jittering" with the bottom layer image. When you think you have it right drop it, put a crop box around the bottom layer checking resolution in the info tray, reset the top layer to 100% (0 translecency), crop, save over the original frame.
To return for the next frame, undo the crop, delete the top frame and begin again.
Make no mistake this is INCREADIBLY time consuming, and you will (eventually) learn a lot of keyboard shortcuts and be screaming at the authors for not including more. I use Animation Shop to Save frames as indexed compressed .tga files and re-producing them into an AVI but other programs will suffice. As I'm sure Adobe Photoshop or so would do in place of PaintShop Pro. As I say I could use AnimationShop to re-produce the AVI.
The De-Jitter algorythems are based on the complex image recognition technologies being developed for f.e. Face Recognitian, still in their infancy, but essential to match one image to another regardless of their angle or rotation, the shadow cast upon them etc.
To be achieved correctly the computer has to have some idea of the three-dimentional properties of the object based on it's two dimentional representation, and be able to say "if it was lighted differently, (moves under a shadow) rotated differently etc would this object look like this".
The De-Jitter also needs to know what the subject of your clip is. Dose it De-Jitter the fishermans head which occupies 20% of the screen, the boat behind him which occupies 60% of the screen or the horrizon behind that which occupies the rest.
This level of complexity is not cheep. And of course with current processing power even at 3GHz etc we need to cheat and drop/fake some of the processes mentioned above.  The need exists however and I have every confidence that it 3-5 Yrs this technology will be common place. MPEG 4 technology is comming close to including similar practices in the CoDec alone.  |
|
| uwek |
| Posted: Sep 21 2002, 02:55 AM |
|
|
Unregistered
|
Mayby the following concept might give a good and easy to implement dejitter filter:
When the filter is configured, the user has to open the preview. Now he has to select in the preview a reference frame and press in the filter config a 'pick reference frame' button. By sampling the actual frame, the filter can get the actual frame as preview. Now the user has to go to next frame in the preview. The filter must now show the next frame and the reference frame each 50% transparent over eachother. The user could now adjust rotation and translation of the actual frame by pressing 'adjustment' buttons (Up, down, left, rigth, rotate cw/acw) on the config dialog. After each buttonpress the filter must call the RedoFrame method to update the actual frame (still overblended with the reference frame, but transformed according to the chossen adjustment). When the user is satisfied with the result, he can select the next frame to adjust.
There is still a lot of handwork left, but might be much easier then processing each frame in a picture-tool.
The most compicated thing in such a filter would be a nice image transformation (bicube or so...) which runs fast and produces good results.
If there would be such a translation library for free (which works with VirtualDub VFBitmaps, I would be able to implement the rest of the filter, I there is still interest... |
|
|
