lolishinshi/imsearch

基于特征点匹配的大规模相似图片搜索

What it solves

imsearch is designed to find a full image within a large-scale dataset using only a small cropped screenshot of that image. It solves the problem of identifying the original source image when only a partial view is available.

How it works

The tool uses feature point matching to identify similarities. It leverages the FAISS library for efficient indexing and searching of these feature points. The process involves:

  1. Adding Images: Scanning directories for image files (jpg, png, webp).
  2. Training: Creating a BIVF index by training a clusterer on a representative sample of the image data to ensure balanced search speeds.
  3. Building: Constructing the final index based on the trained quantizer.
  4. Searching: Matching the feature points of a query image against the index to find the most similar images.

Who it’s for

Users who need to perform large-scale reverse image searches, specifically those looking to locate full images from small fragments or screenshots.

Highlights

  • Large-scale capability: Optimized for handling datasets ranging from a few thousand to millions of images.
  • Feature-point based: Specifically tuned for partial image matching rather than general image similarity.
  • HTTP API: Includes a built-in server mode that can load the index into memory for high-performance API access.
  • Docker support: Provided for easy deployment.

Related

  • Dispatch
    Image Similarity with Hugging Face Datasets and Transformers

    Hugging Face demonstrates how to build an image similarity system using the Transformers and Datasets libraries by computing dense vector embeddings and measuring cosine similarity.

  • Dispatch
    Image search with Hugging Face datasets

    Hugging Face demonstrates how to build an image search application by combining the datasets library, CLIP embeddings, and FAISS for efficient similarity search.

  • Project
    facebookresearch/faiss

    A library for efficient similarity search and clustering of dense vectors, capable of handling billion-scale datasets using CPU and GPU acceleration.

  • Project
    nachifur/MulimgViewer

    A multi-image viewer designed for efficient comparison, filtering, and stitching of large image datasets, featuring parallel zooming and automated figure generation.

  • Project
    TechyNilesh/DeepImageSearch

    A Python library for building AI-powered image search systems supporting text, image, and hybrid search using multimodal embeddings and vector indexing.