This adds a new palette sorting algorithm that attempts to minimise
entropy by an approximate solution to the Traveling Salesman Problem.
The algorithm comes from "An efficient Re-indexing algorithm for
color-mapped images" by Battiato et al
(https://ieeexplore.ieee.org/document/1344033).
It's fast and effective and works in addition to the luma sort (which
remains the single most effective sort). In order to keep lower presets
fast though, I've only enabled this for o3 and higher.
Results on a set of 190 indexed images at `-o5`:
18,932,727 bytes - master
18,578,306 bytes - PR
18,559,863 bytes - PR + #509
(These images may be particularly suited to alternative sorting methods
- the gains here are not necessarily what should be expected on average)
Note I looked into the 120 different palette sorting methods from
TruePNG, as mentioned in #74 (and seen in action in the Zopfli KrzYmod
fork). They're... largely ineffective. The combination of all 120
methods are outperformed by just the existing luma sort plus this new
one. That's not to say there's nothing further to be gained from them,
but trying to brute force all the combinations definitely seems like a
bad idea. There are other algorithms I hope to explore in future...
@ace-dent Thought this might interest you
UPDATE: I realised a quick tweak to alpha values in the luma sort can
provide a great improvement on images with transparency. The following
numbers were taken with PR #509 as base.
`-o2`:
19,065,549 bytes - base (luma sort)
18,949,747 bytes - modified luma sort
`-o5`:
18,922,165 bytes - base (luma sort)
18,559,863 bytes - new sorting algorithm + luma sort
18,544,813 bytes - new sorting algorithm + modified luma sort
This PR makes the oxipng binary process multiple files in parallel,
finally fulfilling #275. There seemed to be some debate about whether
oxipng _should_ do this or not but there's a couple of reasons I think
it makes sense:
1. The concern seemed mostly around the complexity of such a feature.
Not to worry, it was trivial* 🙂
2. Since then, oxipng has dropped from a max of something like 180
simultaneous compression trials down to 10, which is very much a good
thing but it does mean it's not utilising any more cores than that.
Some benchmarks on around 100 files on a machine with 8 cores:
Level | Master time | PR time
-|-|-
2 | 28.303 | 19.005
3 | 36.507 | 23.089
5 | 1:10.86 | 1:16.01
*Some additional changes were required in order to make sure sensible
output is printed to the terminal, since things won't be in order
anymore. Here's some example output from before:
```
Processing: tests/files/fully_optimized.png
file size = 67 bytes (0 bytes = 0.00% decrease)
File already optimized
Processing: tests/files/corrupted_header.png
Invalid PNG header detected
Processing: tests/files/verbose_mode.png
file size = 102480 bytes (12228 bytes = 10.66% decrease)
Output: tests/files/verbose_mode.png
```
And after:
```
Processing: tests/files/verbose_mode.png
Processing: tests/files/fully_optimized.png
Processing: tests/files/corrupted_header.png
tests/files/corrupted_header.png: Invalid PNG header detected
tests/files/fully_optimized.png: Could not optimize further, no change written
102480 bytes (10.66% smaller): tests/files/verbose_mode.png
```
Closes#275, #84, #169, #196 and #419.
[edit] This is the last thing I wanted to land before the next release 🥳
* Add .whitesource configuration file
* Experimental: allow Zopfli to use any size BufWriter
* Allow user to specify the output buffer size as well
* Allow user to specify maximum block splits
* Reformat and fix warnings
* Use deflater on iCCP chunk as well
* Bug fix: need to implement Zlib format
* Make functions const when possible
* Switch to using zopfli::Options in prep for https://github.com/zopfli-rs/zopfli/pull/21
* Switch to using zopfli::Options in prep for https://github.com/zopfli-rs/zopfli/pull/21
* Cargo fmt
* Fix compilation
* Fix tests
* Fix more lints
* Fix more lints
* Fix compilation more
---------
Co-authored-by: mend-bolt-for-github[bot] <42819689+mend-bolt-for-github[bot]@users.noreply.github.com>
Co-authored-by: Chris Hennick <hennickc@amazon.com>
Co-authored-by: Chris Hennick <4961925+Pr0methean@users.noreply.github.com>
* Keep track of number of pixels in each scanline
* Expand depth to 8-bit
* Attempt expand bit depth
* Simplify tracking of reduction_occurred
* Fix test
* Simplify depth handling in reductions
* Add tests for disabled reductions
* Fix bKGD conversion from gray to palette
* Allow grayscale reduction from 16 to 4 or less
* Refactor reduction evaluation sequence
* Separate palette into new file
* Refactor ColorType
Move transparency and palette data into the ColorType
* Fixup tests
* Make more use of helper functions
* Change BitDepth to u8 representation with TryFrom
* Fix clippy lints
* Don't use unstable language features
* Restore documentation on transparency/palette
* Make dependency on `image` optional
After PR https://github.com/shssoichiro/oxipng/pull/481 was merged, the
`image` dependency became unused when building with debug assertions
disabled, as it is only used to implement output sanity checks when such
assertions are enabled.
The `image` crate transitively pulls a significant amount of
dependencies, so it's useful for OxiPNG users to get rid of them when
not needed.
[Cargo does not allow specifying dependencies that are only pulled when
debug assertions are
enabled](https://github.com/rust-lang/cargo/issues/7634), so the next
best way to give users some flexibility is to gate those debug
assertions behind a feature flag.
These changes add a `sanity-checks` feature flag that controls whether
the `image` crate and the related sanity checks are compiled in. This
feature is enabled by default to keep debug builds useful to catch
problems during development.
* Fix Clippy lints
* Run tests with new sanity-checks feature enabled
* Refactor filters as enum
* Include filter byte in filter output
* Add entropy filter
* Add bigrams filter
* Add bigram entropy filter
* Add brute filter
* Replace bit-vec
* Add tests and benches
* Show filters in help
* Use FxHasher in color to palette
* Use windows function for minor improvement
* Switch main compressor to libdeflate
* Use libdeflater in evaluate
* Use libdeflater to inflate
* Use libdeflater crc
* Tidy up
* Fix benches
* Allow libdeflater/freestanding feature
* Fix building without zopfli
* Update and optimize dependencies
These changes update the dependencies to their latest versions, fixing
some known issues that prevented doing so in the first place.
In addition, the direct dependency on byteorder was dropped in favor
of stdlib functions that have been stabilized for some time in Rust, and
the transitive dependency on chrono, pulled by stderrlog, was also
dropped, which had been affected by security issues and improperly
maintained in the past:
- https://github.com/cardoe/stderrlog-rs/issues/31
- https://www.reddit.com/r/rust/comments/ts84n4/chrono_or_time_03/
* Run rustfmt
* Bump MSRV to 1.56.1
Updating to this patch version should not be cumbersome for end-users,
and it is required by a transitive dependency.
* Bump MSRV to 1.57.0
os_str_bytes requires it.
* Add initial support for changing Zopfli iterations
PR https://github.com/shssoichiro/oxipng/pull/445 did some dependency
updates, which included using the latest zopfli version. The latest
version of this crate exposes new options in its API that allow users to
choose the desired number of Zopfli compression iterations, which
may greatly affect execution time. In fact, other optimizers such as
zopflipng dynamically select this number depending on the input file
size (see: https://github.com/shssoichiro/oxipng/issues/414).
As a first step towards making OxiPNG deal with Zopfli better, let's add
the necessary options for libraries to be able to choose the number of
iterations. This number is still fixed to 15 as before when using the
CLI.
* Fix Clippy lint
Co-authored-by: Josh Holmer <jholmer.in@gmail.com>
* Update and optimize dependencies
These changes update the dependencies to their latest versions, fixing
some known issues that prevented doing so in the first place.
In addition, the direct dependency on byteorder was dropped in favor
of stdlib functions that have been stabilized for some time in Rust, and
the transitive dependency on chrono, pulled by stderrlog, was also
dropped, which had been affected by security issues and improperly
maintained in the past:
- https://github.com/cardoe/stderrlog-rs/issues/31
- https://www.reddit.com/r/rust/comments/ts84n4/chrono_or_time_03/
* Run rustfmt
* Bump MSRV to 1.56.1
Updating to this patch version should not be cumbersome for end-users,
and it is required by a transitive dependency.
* Bump MSRV to 1.57.0
os_str_bytes requires it.
The for loop in apply_preset_5 was not including 9.
apply_preset_6 now builds now calls apply_preset_4 instead of
apply_preset_5, and adds all compression levels from 1 to 9.
Co-authored-by: Nino Burini <nburini@jabra.com>
For RGB(A) images that contain gray colors, this reduction can achieve
significant space savings. However, in the absence of gamma correction
data, some PNG decoders assume more exotic color spaces for grayscale
images instead of the ubiquitous sRGB. This results in gamma
miscorrection, and for the end user this means that colors will look
wrong, like "washed-out". Java's ImageIO class, which is popular in the
JVM world to read PNG files, uses rather unconventional defaults, as
explained in this StackOverflow question: https://stackoverflow.com/questions/31312645/java-imageio-grayscale-png-issue
Gamma miscorrection problems aside, OxiPNG currently tries hard to
reduce RGB(A) images to grayscale, because it expects that reduction to
be quite effective. However, in some cases, OxiPNG generates smaller
PNG files when reducing grasycale RGB(A) images to paletted color than
actual grayscale color. For example, let's say that "~/gray.png" is a
256x256 RGBA image entirely filled with (119, 119, 119, 255) pixels.
OxiPNG, by default, reduces this image to grayscale and achieves a
68.23% decrease:
$ cargo build --release && target/release/oxipng -omax --out ~/out.png ~/gray.png
Processing: /home/user/gray.png
256x256 pixels, PNG format
4x8 bits/pixel, RGBA
IDAT size = 604 bytes
File size = 661 bytes
Reducing image to 1x4 bits/pixel, Grayscale
Trying: 144 combinations
Found better combination:
zc = 6 zs = 0 f = 0 153 bytes
IDAT size = 153 bytes (451 bytes decrease)
file size = 210 bytes (451 bytes = 68.23% decrease)
Output: /home/user/out.png
However, if the --ng option that this commit adds is used to skip the
grayscale reduction step, OxiPNG reduces to a single color palette
instead, which is much more efficient, achieving a 84.42% decrease:
$ cargo build --release && target/release/oxipng -omax --ng --out ~/out.png ~/gray.png
Processing: /home/alejandro/gray.png
256x256 pixels, PNG format
4x8 bits/pixel, RGBA
IDAT size = 604 bytes
File size = 661 bytes
Reducing image to 1 bits/pixel, 1 colors in palette
Trying: 144 combinations
Found better combination:
zc = 3 zs = 3 f = 0 31 bytes
IDAT size = 31 bytes (573 bytes decrease)
file size = 103 bytes (558 bytes = 84.42% decrease)
Output: /home/alejandro/out.png
While OxiPNG should arguably be made smarter to better handle these
cases, in the meantime, adding an option to manually skip that grayscale
reduction can't hurt. In fact, it may even help users achieving the most
out of current versions of OxiPNG, and developers reasoning about what
makes a grayscale-like RGB(A) image compress better with a color
palette.
Due to the reasons stated above, this adds a simple "grayscale_reduction"
option to the Options struct, and a "no-grayscale-reduction" command
line switch, that makes OxiPNG not try this problematic grayscale
reduction on RGB(A) images.
* Switch to crossbeam-channel + rayon::spawn
* Remove thread_spawn for evaluation altogether
This allows to avoid a deadlock when there is only one Rayon thread, and doesn't sacrifice performance, since the caller of .get_result() had to always block on the iterator to be finished anyway, and all the messages are already sent from separate threads.
* Fix `verbose_mode` test
This one is easier to "fix", since we're in control of the rayon pool - just adding one extra thread to the default number to make sure we always can one root "spawn" call without stealing from the actual pool.
* Update --help to exclude -a
* Add a deprecation warning to level 4 constructor
* Initialise logger earlier
* Add warning for level > 3 for non-zlib
It's not obvious immediately that these levels don't have any effect on libdeflater and Zopfli, since they don't iterate over zlib-specific fine-tuned options.
Hence, show warning so that user knows they're getting "downgraded" to level 3.
* Add "max" level alias; more level warnings
* Update --help trial numbers for non-zlib
* Fix incorrect trial numbers
This allows to configure or compile away logging in the library from a single place in Rust apps.
For the CLI side, the usage and output remained the same, except it's now colour-coded.
Fixes#217.
Follow-up to #210.
I haven't noticed / forgotten that clap has own mechanism for conflicts between arguments, and it's probably best to use it instead of custom checks.
Rayon uses a singleton global pool.
By default it's set to a regular spawn handler with number of logical CPU cores, but it can be overridden by Rust applications to customize number of threads, spawn handlers, exit handlers and other options.
Such customization should be usually done at the app level, because if a single library initialises the global pool, then Rayon will prevent any further overrides and they will error out. This can cause conflicts between libraries or library and user code and make them impossible to use together.
Hence, I've removed the `threads` option from the `Options` struct and instead moved initialisation to the CLI part of the codebase (main.rs).
Users of the library that didn't depend on custom `threads` number can keep using it as before - they'll still get same number of threads as number of logical CPU cores, while users who need fine-tuning, can do that by customizing rayon pool themselves at the top level of the app.
Note: another alternative to keep the option could've been to use `ThreadPoolBuilder::build` + `ThreadPool::install` to use a local pool just within OxiPNG, but that would ignore any customizations made by users in top-level pool and would prevent usage on targets that require custom spawn handlers like WebAssembly. As such, I've decided to avoid it.
Co-authored-by: Josh Holmer <jholmer.in@gmail.com>
In order to avoid problems with mutable references,
the approach is to use the eval sender for non-parallel.
Computations will still be done synchronously
with the fake synchronous rayon module.
* Skip baselines earlier in the evaluator
We know that we'll throw them away anyway, so there is no point in even sending them to Sender and comparing them with others once we used them for the best candidate size.
* Simplify comparison logic
* Improve performance consistency
Switch from HashMap / HashSet to IndexMap / IndexSet for consistent iteration order of various options and, as a result, more predictable performance.
