04_MA Forum video


I made the video you see above by:

1) filming Nicola walking in circles
2) choosing a circle where the beginning and end steps were lined up
3) this turned out to be 169 frames
4) removed 2 frames to end up with a prime number of frames
5) wrote a java program to create an AviSynth file to re-order the frames
6) opened AviSynth file in VirtualDub
7) export as an AVI file.

AviSynth
This is a frame server which can send a series of frames (or AVI clips) to film editing software. The frames are listed in a text file. The editing software is fooled into thinking the text file is actually an AVI file, and plays it as a video. The java program outputs the text file algorithmically as a huge list of frames. First each frame is sent in sequential order, then every other frame, then every third frame…and so on until every 167th frame is sent (which is effectively a still image). Here is an example, the second line from the text file:

clip2=Trim(41,41) ++ Trim(43,43) ++ Trim(45,45) ++ Trim(47,47) ++ Trim(49,49) ++ Trim(51,51) ++ Trim(53,53) ++ Trim(55,55) ++ Trim(57,57) ++ Trim(59,59) ++ Trim(61,61) ++ Trim(63,63) ++ Trim(65,65) ++ Trim(67,67) ++ Trim(69,69) ++ Trim(71,71) ++ Trim(73,73) ++ Trim(75,75) ++ Trim(77,77) ++ Trim(79,79) ++ Trim(81,81) ++ Trim(83,83) ++ Trim(85,85) ++ Trim(87,87) ++ Trim(89,89) ++ Trim(91,91) ++ Trim(93,93) ++ Trim(95,95) ++ Trim(97,97) ++ Trim(99,99) ++ Trim(101,101) ++ Trim(103,103) ++ Trim(105,105) ++ Trim(107,107) ++ Trim(109,109) ++ Trim(111,111) ++ Trim(113,113) ++ Trim(115,115) ++ Trim(117,117) ++ Trim(119,119)

It would have been possible to make this video using traditional (non algorithmic) editing techniques, but each frame would have to be edited in place by hand, and there are over 3500 of them for the two minute film – which is a lot of editing. I also need to look into ways of ghosting (trace i think its called or motion blurring) films in Final Cut Pro, which also has its own scripting language FXScript – this might save a lot of mathematical grief.

Spent several days trying to develop this study in other directions. What i wanted was to fill out the circle by blending each woman with several ghost images. I was thinking to generate a more solid form. Nothing worked very well so i stuck with the original and added a soundtrack. Here are some things that didn’t work:

n0.jpg

n1.jpg

n2.jpg

To get the blending of ghosts i wanted i decided to work with just the women without the backgrounds. Doing it this way immediately caused problems because the threshold used to isolate the woman is difficult to find. Green screening would help here but then you lose the slight changes in the background which locate the woman in space as opposed to pasted on top of it. Also the shadows around the feet were a headache.

n3.jpg

Ideally i thought i’d use every single frame in the ghost images – graduating the opacity of them as they got further away from the central woman. The problem here was that the more images were on top of each other the more likely it was that the definition of the woman’s shape was lost – because the decision whether to show the ghost pixels or not was done by assessing whether the ghost RGBcolor was larger than the latest woman RGBcolor. This maximum was done by comparing the sum of the squares of the red,green & blue components as you add each new ghost. This is rather crude.

blur1.gif

Better was to have a gap between ghosts (this refers back to the Continuous vs. Discrete aesthetic i made in my original project proposal schematic):

blur2.gif

The final set of problems came when trying to posistion these BMPs back into the yard scene.

n4.jpg

The lesser ghosts carry with them a lot of very dark color which makes a mess unless dealt with. this meant i had to have some crude algorithm like so:
1) if the ghost pixel is colorful enough then use that pixel rather than the background pixel
2) if the ghost pixel is too dark then use a ratio between the ghost and background pixels

n5.jpg

This worked to an extent but the image bacame much lighter than it should be. And as a film sequence it was nasty.

In the end i decide none of these tests were as strong as the original MA Forum video & I had ran out of time. I will try to sort out these problems when i’ve got a better understanding of alpha blending. I think that might be the right way of solving this – by not removing the background in the first place.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: