nortti OH1CAU

Remembering "Wiznaibus"

FFT's original localization is amazing because it's clearly written by native speakers who actually do know how to write English well, but also produced some of the most amazing misromanizations I've ever seen

I was looking at deinterlacing filters and saw the term "laplacian" come up again. I wondered to myself, "what does that mean anyway, and how does it relate to video processing?"

So I said, "wikipedia, explain what laplacian means" and upon reading the first paragraph felt like I'd been drag-and-dropped into the middle of the Pacific ocean, left to fend for myself. This sounds more like Star Trek technobabble than Star Trek technobabble. This reads like the exaggerated parody of Star Trek technobabble. I lowkey love it. It's like peeking into a Pandora's box of a whole other side of human interest and then gently closing the lid again going "I'm not going down that rabbit hole."

wasn't that one of the spears in Evangelion or something.

brb gotta convolve my Gaussian so it doesn't turn into a dark blob. Gotta avoid those dark blobs, y'know. That's why I never go out without my multi-scale blob detector with automatic scale selection. Thank god for that scale-normalized Laplacian operator or I don't know how we'd all still be alive.

I just wanted to learn more about deinterlacing.

So I would strongly encourage you to start here to make some sense about what's going on here:

In a high-level explanation of what's going on here is that you're basically doing an operation in frequency space -- any time you're doing a convolution like that, you're really doing something akin to a band-pass filter. If you know what a Fourier Transform is (this is probably going to show up in the video), a convolution on a piece of data or an image or signal is the same as a more basic operation (multiplication) in the Fourier (frequency space) representation. In this case what we're talking about with 'blob detection' is sort of like a high-frequency filter; you're effectively marking regions of space that are all pretty similar to each other. This is more-or-less how edge detection works in any sort of computer or machine vision.

When they're talking about blobs it's not just the edge but basically the filled in area the edges encompass. By taking the Gaussian you're blurring the image so that areas that are similar and close to each other become more similar, by averaging a range of pixels around a point. This means that there is less change between adjacent pixels and so means that the image becomes lower-frequency; it's a low-pass filter, applied to a 2D signal.

I'm more familiar with the Laplacian in terms of physics, esp. classical mechanics. It's a way of representing a function in terms of its second derivative (differential equations, yay!). It shows up a lot in classical physics because classical physics measures things in terms of forces, which are proportional to acceleration, which is the second derivative of motion. That this has direct ties to the roots quantum mechanics is a key part of why quantum mechanics is the way it is, by the way. The Hamiltonian operator in quantum mechanics is the way that the laplacian relates to the energy in the system; because we're doing this analysis on frequency, all the fun limitations of wavelet transforms start applying to the solutions our physics calculations give us, where analysis of a wide region of time gives us a very good sense of the frequency of a signal but a bad notion of an immediate signal value, and a short region of time gives us a very good idea of what a signal value is but not what its frequency is. In terms of quantum particles these are directly related to momentum and position. This is getting a bit outside of your image processing context but I wanted to just explain a bit about where my experience is, as it's slightly outside where you're coming from.

What you're measuring with the Laplacian is regions where the rate of change is either high or low. As a result, you're basically using it to draw boundaries between the regions of similar color (which are more similar now because of the Gaussian step you applied -- you removed noise from the data). What you're left with is boundaries between similar regions of what I've been presuming in writing this to be brightness, which lets you identify where your 'blobs' are.

I presume you're aware of what a kernel is (or at least the video should help explain it) in terms of image processing operations, in which case this also may help understand some things: https://en.wikipedia.org/wiki/Discrete_Laplace_operator

Anyway the reason that all of this is relevant to deinterlacing is that you're trying to detect how similar a specific image is to another one by using this blob calculation across multiple frames (or fields) and seeing how it all lines up. If you try to merge two interlaced fields into a single frame when there's high motion you get a particularly nasty blur effect. At an obnoxiously high level: When motion is low, you can merge the frames without a lot of compensation and get a frame that has an effective vertical resolution that's more reflective of the frame rather than the field, but when motion is high you'll want to sacrifice the resolution to cut down on that blur.

My first gaming PC couldn't run XP but it could run ME. and that crashed constantly. it sucked so bad. oh my god. fuck.

Shoutout to Windows 2000, which I kept using for years until a driver conflict with my brand new graphics card caused it to gradually corrupt the hard drive over a 72-hour period, ultimately resulting in four format-and-reinstall-everything ordeals in a row before the problem was finally diagnosed.

I used windows 2000 until late 00s, when my hard drive crashed and I decided to go full-time linux. it was really rock solid, and save for lack of modern internet software still surprisingly usable today when run in a virtual machine

Anakin Skywalker was a legendary Force-sensitive human male who was a Jedi Knight of the Galactic Republic and the prophesied Chosen One of the Jedi Order, destined to bring balance to the Force.

YEAH? THAT'S WHAT HE WAS CALLED? ARE YOU SURE HE DIDN'T GO BY ANOTHER NAME? A PERHAPS SLIGHTLY MORE FAMOUS NAME? A NAME UNDER WHICH HE MAYBE DID SOME OTHER MORE HISTORICALLY SIGNIFICANT THINGS?

Anakin Skywalker was a legendary Tatooine-born human male pod racer who won the Boonta Eve Classic at the record setting age of just 9 years old. After his retirement, he joined a monastic order, and later became involved in politics.