pure-sift

pure-sift is a comprehensive computer vision engine developed entirely in Rust, emphasizing safety and performance with a commitment to zero-unsafe code (#![forbid(unsafe_code)]). The project implements the Scale-Invariant Feature Transform (SIFT), along with robust alignment techniques through RANSAC (RANdom SAmple Consensus) for both 2D similarity and planar homography alignment. With no dependencies on C/C++ FFI or OpenCV/CMake/LLVM, this engine compiles directly into static binaries suitable for a variety of environments including Docker, WebAssembly, and mobile targets.

Key Features

• SIFT Feature Extraction: Implement effective feature extraction using Scale-Space Difference-of-Gaussians keypoints, comprehensive descriptor extraction, and adaptive orientation assignments.
• Robust Matching and Alignment: Leverage Lowe's ratio matching alongside horizon-guarded homography estimation to achieve precise geometric transformations.
• Independent Visual Catalog Search: Utilize fast Hamming shortlisting coupled with geometric verification for efficient retrieval of similar images from a catalog.
• Command-Line Interface (CLI) and Web-based Interface: Includes a drop-in API powered by Axum with a visual interface for easy access to functionalities, such as viewing keypoints and aligning images.

Example Usage

This library can be seamlessly integrated into Rust projects. Here’s how to extract SIFT features:

use image::ImageReader;
use pure_sift::{extract_sift_features, match_descriptors, estimate_homography_ransac};

let q_luma = ImageReader::open("query.jpg")?.decode()?.into_luma8();
let t_luma = ImageReader::open("target.jpg")?.decode()?.into_luma8();

let q_feats = extract_sift_features(&luma_to_array2(&q_luma), 3, 3, DEFAULT_CONTRAST_THRESHOLD);
let t_feats = extract_sift_features(&luma_to_array2(&t_luma), 3, 3, DEFAULT_CONTRAST_THRESHOLD);

let matches = match_descriptors(&q_feats, &t_feats, DEFAULT_MATCH_RATIO);
let (inliers, homography) = estimate_homography_ransac(&matches, DEFAULT_RANSAC_REPROJ_THRESHOLD);

HTTP API Endpoints

• GET /: Access an interactive SIFT and RANSAC workbench.
• POST /api/v1/match-features: Match and process features between images efficiently.

Performance Metrics

The engine has been benchmarked on real-world data, delivering consistent high-quality results across varying input scenarios. Detailed specifications and mathematical derivations are included in the project documentation.