Identical Image Search System (ISS)

The Identical Image Search System (ISS) is a powerful tool designed to search for identical images using a pretrained model called SuperGlobal. It utilizes the FAISS library in Python for efficient storage and retrieval of image embeddings, and SQL for managing metadata.

Table of Contents

1. Installation
2. Usage
3. Model
4. Database
5. Examples
6. License

Installation

To get started with ISS, follow these steps:

1. Clone the repository: git clone https://github.com/Shintifo/ISS.git
2. Navigate into the project directory: cd ISS
3. Install the required packages: pip install -r requirements.txt

Usage

To use ISS, you need to have your images ready in a datasets/{dataset_name} directory. Also, you need to create folder databases. Firstly, you need to initialize the database with images. python database.py dataset_name

Then, you can use the main script to search for identical images. python search.py dataset_name query_image

After, it display all similar images by set threshold (threshold can be changed in run.py)

Model

ISS uses the SuperGlobal pretrained model to generate image embeddings. This model was converted to ONNX format and executed in run.py.

Database

• SQL: ISS uses SQL to store and manage metadata about the images. This includes image by itself and image name.
• FAISS: ISS uses index based database to store vector embeddings.

Examples

Query:

Search Result:

License

ISS is released under the MIT License. See the LICENSE file for more details.