2 releases
0.1.1 | Dec 4, 2024 |
---|---|
0.1.0 | Dec 4, 2024 |
#162 in Science
227 downloads per month
19KB
261 lines
is-photo
Utility to determine if an image is likely a photograph or a 2D graphic, such as a logo, illustration, or digital art.
It does this by taking various statistics from the image and running them through a pre-trained logistic regression model, along with a few sure-fire heuristics.
On my test set of around 1500 images, it has a 94% accuracy rate. Feel free to submit links to additional image sets to train on in an issue!
Example
# fn main() -> Result<(), Box<dyn std::error::Error>> {
let img = image::open("test.jpg")?;
let analysis = is_photo::analyze(&img).expect("Failed to analyze image");
let is_photo = analysis.is_photo(&is_photo::STANDARD_MODEL);
# Ok(()) }
Future work may include training on a larger dataset, and possibly using a full neural network instead of logistic regression.
Dependencies
~5–9.5MB
~103K SLoC