Welcome to the Rust Amazon Captcha Solver, an tool designed to resolve Amazon.com captchas without relying on Tesseract OCR or any external OCR software. Inspired by the exceptional work of gopkg-dev/amazoncaptcha and a-maliarov/amazoncaptcha, this project builds upon their dataset, presenting it in an uncompressed bincode format.
Amazon captchas exhibit a repetitive nature, with characters consistently less than 32 pixels in size, resulting in a finite number of possible combinations. We leverage the dataset.bin containing most of these possibilities. When a character doesn't match exactly, we employ a similarity comparison function for accurate resolution.
```rust use amazoncaptchars::new_solver;
let image = image::open("captcha.jpg").unwrap();
let solver = newsolver(); let response = solver.resolveimage(&image).unwrap();
assert_eq!(response, "caggpa"); ```
Starting from version 0.2.1, we've introduced a benchmarking feature to assess resolved captchas directly on Amazon. We conducted over 2000 tests with the following results:
Resolved: 1990/2020
Precision: 98.51%
You can run the benchmark using the following command:
cargo run -r --example benchmark_infinite
Additionally, we have a second benchmark performed on our proprietary labeled dataset located in example/dataset. This dataset contains only 100 images and yielded the following results:
Solved: 97/99
Precision: 97.98%
Average resolve time: 6.76ms
Total time: 0.75s
Execute this benchmark with the following command:
cargo run -r --example benchmark