Enum Quantize 

pub enum Quantize {
    Lossless,
    Step(f64),
    NoiseLevel(f64),
}

How a floating point image’s values are turned into the integers a compressor can work on.

Quantising is what makes a floating point image compress at all well, and it is lossy: what comes back is within one step of what went in. The step is the choice being made here.

Variants

Lossless

Do not quantise. The values are compressed as they stand, which only Compression::Gzip, Compression::ShuffledGzip and Compression::None can do, and nothing is lost.

Step(f64)

Quantise in steps of this many units of the image’s own values.

NoiseLevel(f64)

Quantise in steps of the tile’s own estimated noise divided by this.

Four is the usual choice: it keeps the quantisation step well below the noise already in the data, so nothing measurable is lost, while throwing away the low bits that are noise anyway and would otherwise compress terribly.

The noise is estimated from the median absolute third-order difference between neighbouring pixels, which is what the convention’s own reference implementation does. That implementation takes the smallest of three such estimates rather than this one alone, so it may settle on a slightly different step for the same tile; either is a fair reading of “the noise in this tile”.

Trait Implementations

impl Clone for Quantize

fn clone(&self) -> Quantize

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for Quantize

impl Debug for Quantize

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for Quantize

fn default() -> Quantize

Returns the “default value” for a type.

impl PartialEq for Quantize

fn eq(&self, other: &Quantize) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for Quantize