Consider adding "blur" effect

[ericrosenbaum]

We should consider adding a blur effect to the graphic effects menu.

Questions:

• Would a simple "box blur", applied iteratively as the effect value increases, be fast enough?
• The sprite's graphical bounds won't change, so the blur will have an edge. Is this okay? Is there a workaround (i.e. should the blur make the sprite larger?)
• Negative blur has no effect I guess?

(similar to https://github.com/LLK/scratch-gui/issues/683)

[mrjacobbloom]

If you do decide to increase the bounds on blur, you may be able to use an outset algorithm instead of recalculating the whole hull

[Clark-E]

I’ve worked with blur once before in web browsers (using css filters). It isn’t too slow, as long as you’re don’t make the blur crazy big and you’re not recalculating every frame.

But, the problem lies in the fact that a user could potentially blur a really fast animation with many frames. That would take a while to compute.

That said, you should take my input with a grain of salt. Maybe WebGL can handle blur really well.

[adroitwhiz]

I'm a big fan of this idea. The blur effect seems really useful--at the very least, more useful than, say, "whirl" or "fisheye". Animation projects in particular could use it as a way to simulate camera focus.

• The sprite's bounds can be easily increased GPU-side via changing gl_Position in the vertex shader. This may be significantly more complex to implement for silhouettes, however.
• A box blur would likely be the best option for speed and appearance. A Gaussian blur lacks the "bokeh-ness" that animation projects would greatly benefit from. Circular separable convolution would look even more bokeh-ish but requires way more computational resources.
• A two-pass blur is a lot faster than a one-pass blur for larger blurs, but that would require implementing a two-pass effect system.
• Larger blurs would probably start to lag out on lower-end GPUs. Either the blur could be clamped to a certain size, or large blurs could operate on a downsampled texture (which would add yet more complexity).

@cwillisf thoughts on this?