tv1d's Introduction

`tv1d`

Total variation denoising algorithms for 1d data.

Total Variation Denoising

Total variation (TV) denoising removes fine, potentially spurious noise from a discrete signal while preserving its sharp edges. The technique is well-suited for recovering piecewise constant signals and is often favored for denoising images over alternatives that would smooth edges.

Install

Add this to your Cargo.toml:

[dependencies]
tv1d = "0.1.0"

Example Usage

extern crate tv1d

fn main() {
    let input = vec![13.0, 24.3, 63.41, 13.6];
    let lambda = 3.0;

    let output = tv1d::condat(&input, lambda);
}

Documentation

Read the documentation on Docs.rs.

License

This crate is licensed under MIT license (LICENSE).

Using Rust crate from other languages

Please check out the Rust Book's chapter "Rust Inside Other Languages".

Contribution

See CONTRIBUTING.md!

Denoising Example

Following is a visualization of tv1d::condat with varying lambda applied to Human Death-associated protein 6 (DAXX) transcript variant 1 expression data from UCSC Human Genome database.

"base pair" denotes the position at the transcript. "expression value" denotes the level of expression at a given "base pair" after denoising. "lambda" is the degree of denoising applied to the signals, and 0 is the raw signals.

The plots were generated with ggpy.

tv1d's People

Contributors

Stargazers

Watchers

tv1d's Issues

Make a version of `condat()` that takes an iterator

The Condat algorithm is implemented in a way that it updates the denoised output once a jump occurs (example here). The denoised output is stored in the form of segment_lower_bound or segment_upper_bound throughout the program even when the denoised output is not updated yet.

Therefore, this makes the Condat algorithm a great candidate for taking in an iterator. The denoised value for the input signal would be returned after processing each input signal (possibly at this part).

This will make Condat more useful for denoising a stream of data as Condat can update the denoised signals more instantaneously.

`condat` can subtract or add with overflow

https://github.com/jeenalee/tv1d/blob/master/src/lib.rs#L320

Sample code: (I have not actually tested this code but it seems like it should cause the overflow)

let my_vec: Vec<i8> = vec![127];
tv1d::condat(my_vec, 10);

Due to the add and subtract on those two lines of code not being checked, an overflow may occur. This happened to me when I was using tv1d to denoise some audio.

Recommend Projects

jeenalee / tv1d Goto Github PK