Talk:JPEG

It's probably not DCT, it's flood fill !

You make a balanced flood fill over the source image that aborts when the average of all covered pixels differs with the current pixel by more than a tolerance limit. The covered area is then filled with the average value of all of it's pixels. When you compress the result with RLE or ZIP, you'll see that a lot of redundancy can be won without completely ruining the image. When the resulting image doesn't look that good or photorealistic, you can still interpolate the areas with sinus or cosinus formed shades and it will get a bit better. — Preceding unsigned comment added by 87.143.68.134 (talk) 23:56, 2015 February 20 (UTC)

Wrong sizes in RLE

I believe there is a mistake in part showing the result after RLE. There are no sizes 3 (2^3) even though it is possible to fit only 6 values to sizes 1 and 2 combined (2^1+2^2), but the quantized block has 8 values (1, -1; 2, -2, -3, -4; 5, -6). This is the part:

With this in mind, the sequence from earlier becomes: (0, 2)(-3);(1, 2)(-3);(0, 1)(-2);(0, 2)(-6);(0, 1)(2);(0, 1)(-4);(0, 1)(1);(0, 2)(-3);(0, 1)(1);(0, 1)(1); (0, 2)(5);(0, 1)(1);(0, 1)(2);(0, 1)(-1);(0, 1)(1);(0, 1)(-1);(0, 1)(2);(5, 1)(-1);(0, 1)(-1);(0, 0);

I believe this is the right sequence:

(0, 2)(-3);(1, 2)(-3);(0, 2)(-2);(0, 3)(-6);(0, 2)(2);(0, 2)(-4);(0, 1)(1);(0, 2)(-3);(0, 1)(1);(0, 1)(1); (0, 3)(5);(0, 1)(1);(0, 2)(2);(0, 1)(-1);(0, 1)(1);(0, 1)(-1);(0, 2)(2);(5, 1)(-1);(0, 1)(-1);(0, 0);

Am I right or am I missing something?

——

• I think so. Amplitude category appears to follow $SSSS=\lfloor log_{2}({\text{coefficient}})\rfloor +1$. See table F.2 . The system for calculating this isn't elaborated on very much. --Merukit (talk) 06:40, 16 June 2024 (UTC)

References