/vr/ - Retro Games

>>3905751
You mistakenly claim information is more sparse in the sharp image vs the blurred image. The amount of information needed to describe something depends on your choice of how to describe it. For instance, you used a pixel basis to argue that the information content of a sharp image is less than a blurred image. However, countless image compression routines use sinusoidal bases instead to represent images, because image information is often more sparse in a Fourier-like basis than a pixel basis. In fact, the sharp image would require a huge bandwidth to accurately represent the sharp transition (look up Gibbs phenomenon). Furthermore, your claim that one can perfectly reconstruct the source data from a displayed image is actually a claim about the limited amount of information contained within the (bandlimited) signal, and by saying it's lo-res compared to the display, you are implicitly appealing to the Nyquist Sampling theorem.

> for loss of information to take place, the noise level has to exceed the signal level
This is completely false. A bit can flip in transit even with an SNR > 0 dB. In contrast, even when SNR < -90dB you can reconstruct signals perfectly if you are sophisticated and have something like an encoding scheme going on. Cross-correlation is a wonderful thing.

>>3907729
Intuitively I'd suspect a Gaussian filter to be superior from both a computational and aesthetic point of view (it's a separable filter (read really fucking fast) and isotropic (read smooths the same in all directions, unlike bilinear)).

>>3874658
>muh distorted textures
feels bad man

TFW mo gf because affine mapped