ANTILOG_08July18a

###

ANTILOG_08July18a

###

16:57 2018-07-08

- Yesterday, I ran a simple experiment with digital images and was looking at their sizes in terms of kilobytes; I started with a picture that was all white pixels and compared it to a picture that was Gaussian noise; As predicted, the file size for the Gaussian noise was greater than the file size for the all-white-pixels image;
- A few things weren't clear to me yesterday, but I have since done a little research; The images are bigger in terms of bytes than the number of actual bytes or bits that it takes to encode the image; this is because image formats have headers and other metadata, so you end up with a file that is bigger than the theoretical number of bits needed to encode all the pixels with their respective pixel values;
- A standard grayscale image with 256 levels for each pixel requires 8 bits per pixel; An n-bit pixel can take on 2n different values;

Non-monochrome Gaussian noise at 254 kilobytes. A.G. (c) 2018. All Rights Reserved.

- What I was trying to get at was that an image with randomly selected pixel values will be harder to "compress" than an image where there is a lot of "redundancy"; That is to say, the image which has ALL WHITE PIXELS is easy to specific, i.e. a 256 pixel by 256 pixel image with only white pixels could be specified as "W65536", only six characters, the color and the number of pixels with that same value;
- SEE: "Run-length encoding (RLE) is a very simple form of lossless data compression in which runs of data (that is, sequences in which the same data value occurs in many consecutive data elements) are stored as a single data value and count, rather than as the original run.";
- SEE ALSO: "In algorithmic information theory (a subfield of computer science and mathematics), the Kolmogorov complexity of an object, such as a piece of text, is the length of the shortest computer program (in a predetermined programming language) that produces the object as output. It is a measure of the computational resources needed to specify the object, and is also known as descriptive complexity, Kolmogorov–Chaitin complexity, algorithmic complexity, algorithmic entropy, or program-size complexity.";
- So in a sense I was thinking of the "complexity" of the underlying image, and that's what I'm trying to get at; My value of "visual interestingness" or "VI" for short, has something to do with complexity; I'm not yet sure exactly what the correlation is, but I'm working on it, it also has to do with information entropy or "surprisal";
- "Noisy" images have more of an element of surprise, or "uncertainty" if you want to call it that; For VI to occur, you need a certain level of uncertainty, or it will be too redundant, and therefore not with much VI; just like an all-white-pixel image is not "beautiful", it's too simple, it has a very low VI value; But then pure Gaussian noise also has a low VI value, because it's just pure noise; So the magic VI spot is somewhere in the middle, and these visual experiments are all meant to get to a working definition of visual interestingness that can be used to generate visually interesting images algorithmically, or what is called "computational creativity";

Gaussian noise, 109 kilobytes. A.G. (c) 2018. All Rights Reserved.

- "BIT DEPTH is determined by the number of bits used to define each pixel. The greater the bit depth, the greater the number of tones (grayscale or color) that can be represented. Digital images may be produced in black and white (bitonal), grayscale, or color. A bitonal image is represented by pixels consisting of 1 bit each, which can represent two tones (typically black and white), using the values 0 for black and 1 for white or vice versa. A grayscale image is composed of pixels represented by multiple bits of information, typically ranging from 2 to 8 bits or more.";
- "The most predictable image is a large rectangle in a single color. In other words, a scaled-up version of the one-pixel images I discussed in part one. An empty canvas, if you want. A blank sheet. Compression algorithms should be really good at compressing an image where every pixel is the same color. It’s the best-case scenario, the ultimate in predictability – once you’ve seen the first pixel, you’ve seen them all.".