Project 17 ("William")
Short deep-zoom based on a still image from 2004.
This deep-zoom was suggested by William (FractalWizz on YouTube) about a year ago based on a still image I made in 2004. This animation zooms to that still image, then continues into a mini M-set buried a little deeper within it. This is not a very long or very complex animation, but a technological improvement in the colorizing software makes its debut in this project. This is also the first project rendered with anti-aliasing.
The still image that inspired this animation is where the zooming pauses, at time index 1:33, frame 2800. It is the center image in the bottom row of thumbnails below.
|Fast Download||320x240 220 Kbps 4 MB fast start|
|Good Quality||640x480 1.7 Mbps 30 MB fast start|
|High Quality||640x480 10 Mbps 178 MB fast start|
|Fast Download||320x240 200 Kbps 4.3 MB|
The high-quality versions are encoded at 10 Mbps and are very nice, almost indistinguishable from the raw, uncompressed video. As usual, small files are provided for fast download a real-time viewing while downloading. MP4 files are encoded with FastStart so QuickTime should start playing right away.
|Date Generated:||17Jul-9Aug 2009|
|Final Image Size:||6e-40|
|Video Length:||2:00 of fractal, 2:27 overall|
|Rendering Time:||522 hours|
Animations in the cusp region of the Mandelbrot set tend to have a huge amount of visual noise because so many of the image points are very close to the boundary of the set. That causes the pixels to have a sort of sparkling effect that many people find objectionable. The only solution to this problem is to add some sort of visual smoothing filter (that is, a spatial low-pass filter) into the rendering process. This is discussed in detail on the anti-aliasing page in the Technical section. To avoid losing detail from the smoothing, each frame is rendered at a higher resolution than the final video, and then downconverted to the final video size after the smoothing filter is applied.
Project17 is (I think) the first animation published on HPDZ.NET that uses anti-aliasing. It uses 9X oversampling on a regular 3x3 grid and applies a median filter for smoothing. The result is not perfect, and a lot of sparkle noise is still in the video, but it is much better than without filtering.
One of the most challenging aspects of producing high-quality fractal animations is getting the final video colorized well. This is one of the most active areas of research at HPDZ.NET. The challenges have been described on this page in the Technical section.
Project 17 is a demonstration of a new method that looks very promising. A histogram of fractal data is generated based on all the pixels in every frame of the image. That histogram is then used as a color map, and the color map is constant in time. This effectively balances the color usage nicely across the entire video and does so without causing the colors to change with time. The technique needs some further fine-tuning but is very promising.
Although the motion in this video is pretty simple, nothing is ever so simple that it cannot be messed up. This little clip (WMV only, sorry) of the final 27 seconds of the video shows a mistake in coordinating the zooming in with the slight panning motion needed to re-center on the final zoom point. It zooms way too much too early, before the lateral motion begins, so what you see is a 1000X view of the area surrounding the final M-set whipping by at the speed of sound.
The blooper video was a rough draft, made at 320x240 resolution and without any anti-aliasing. You can see that the huge amount of sparkle noise is very distracting. This draft was colored using dynamic adaptive color mapping, so the colors are continually swirling around. This is one of the effects I have been working to eliminate, and the final version with the new global histogram method works much better.