weirdly ruffled top
Made this up last month -- a mock wrap top, using the pretty sheer fabric Shannon gave me for my birthday. I used this pattern: Butterick 3385.
I like it a good bit, except for the weirdly ruffled shoulders. This fabric was just so incredibly hard to work with -- it frayed like crazy, slipped around, and wouldn't hold any folds I pressed into it with an iron. This made hemming and finishing edges really difficult and I totally messed up the ruffles. For the inside seams, I tried french seams for the first time ever, and I must say they turned out nicely :) If I'm feeling productive enough, I may go back and take off the ruffles and just finish the shoulder edges plain instead.
I'll definitely make the pattern again, but next time, I'll use a cotton or something easier to work with.
video time
1. Video is simply a series of images of an event, viewed sequentially in the same space.
2. A multiflash strobe photograph, is a series of images of an event, taken sequentially, and developed on the same piece of paper.
which leads us to ...
3. If you take the time axis of a video and compress it, you can view video in a new way, as a static image or series of images on which all of the time information is represented without having to experience time itself, much like a multiflash strobe photograph.
This is a project I have been mulling over on and off, since the end of 2000, and I really want to get into gear and just do it (especially before somebody else does!) The idea is to take a scene from a video or a series of scenes from a movie, and process the images in the frames so that you can view all of the time information all at once. It would lead to a lot of interesting images -- for example, a conversation would result in still heads with blurry mouths, while a pan would be a streak of city buildings across the image.
The concept below shows an example on the left, compared with a still frame from the same scene on the right.
image credits: strobe photograph taken by students in the strobe project lab of 98. "The Cat Returns" frames are taken from the Studio Ghibli movie, and the project concept heavily photoshopped by myself.
How would this be done?
Language:
I have a variety of choices to program this in.
Java and the JMF libraries are a possibility. I programmed a computer vision project that tracked moving objects in a video using this language before, and it wouldn't be too hard to adapt it. However, I would have to program new modules for outputting the images once they are processed.
A new one is Processing, a language developed specifically for experiementing with graphics, images, and video by Ben Fry and Casey Reas, formerly of the ACG research group at MIT. It looks like a lot of the input/output problems are taken care of, and all I would need to do is write the correct algorithms. I'll need to check it out and see.
Methodology:
Ideally, the final implementation of this project would output a large printable image file with each scene of the processed video separated into its own multiflash-like image. I know it's going to be difficult if not almost impossible to do that right away, so here are the steps I'll follow to get there:
1. Process a scene, and simply average every 5-10 frames in that scene into one image. The result will probably be a big blur that is not helpful to everyone, but it'll be interesting to see what happens.
2. Process a scene, and use a combination of pixel-averaging and computer vision to produce the final image. This will be the brains of the whole operation.
The computer vision aspect will involve finding out what parts of the screen changes from frame to frame, and highlighting that so it doesn't get swallowed into the background. In strobe photography, we mitigate this problem by making sure the background is dark, and the subject is wearing white. We also move the background far away enough from the camera, so that the multiple strobe light flashes won't overexpose it in the eventual developed picture.
In the video processing, I'll probably have to find out what the "background" of an image is by using some sort of difference calculator from frame to frame, and darken it or give it less weight when averaging all frames. In the same vein, I can brighten or give more weight to the foreground -- which should be the part that changes. For part of this, I can reuse an algorithm I coded for my thesis (borrowed from a computer vision paper).
(Another method is to code some sort of tracking algorithm to track objects -- this avoids the problem of things disappearing into the background if it stays still too long, but it's very complicated for the wide variety of objects you'll see in these movies, so I'm not going to do that. Plus, if I write the pixel averaging algorithm right, we should cancel out that problem.)
3. Process a movie, finding where each scene begins and ends, and then processing each individual scene according to step 2 above. This will probably use the same frame difference algorithm as before, and find scenes based on how much of a frame has changed.
4. Arrange all of the scene-images in some aesthetically pleasing pattern. Maybe a long strip of paper, or a poster print.
... and that's it! Now all I have to do is think of a cool name for this project.
ode to the bluebird
Click on image to see full-size. Audio files to follow soon, courtesy of Steve.
