CapSpot 0.1.0

An image matching package using ONNX models

CapSpot

CAPTCHA Resolution Tool

[//]: # ()

Introduction

CapSpot is an image matching library based on ONNX models that locates and identifies sub-images within a target image using a template. It is designed to solve CAPTCHA challenges by finding specific click points in images, making it an invaluable tool for automated testing and security applications.

Example Outputs

Example 1

• Target Image

• Detection Template and Order

• Detection Results

• JSON Response

{'img0': [103, 35, 145, 76], 'img1': [162, 90, 196, 127], 'img2': [175, 189, 199, 218]}

Example 2

• Target Image

  

• Detection Template and Order

  

• Detection Results

• JSON Response

{'img0': [24, 28, 89, 93], 'img1': [243, 60, 300, 119], 'img2': [134, 128, 191, 184]}

Features

• • Utilizes pre-trained ONNX models for object detection and similarity calculations.
• • Supports optional GPU acceleration.
• • Offers a simple and easy-to-use API.

Installation

Via pip:

pip install CapSpot

From source:

git clone https://github.com/miaohancheng/CapSpot.git
cd CapSpot
python setup.py install

Usage

from CapSpot import ImageMatcher

if __name__ == '__main__':
    matcher = ImageMatcher()
    matches = matcher.match_images('./test_pic/test1_mo.png', './test_pic/test1.png')
    matcher.display_results('./test_pic/test1.png', matches)
    print(matches)

Configuration Parameters

•	ImageMatcher class
•	sim_model_name: Name of the similarity model (ONNX format), with a default model provided.
•	det_model_name: Name of the detection model (ONNX format), with a default model provided.
•	match_images method
•	template_file: Path to the template image.
•	target_file: Path to the target image.

Model Downloads

Place the required model files (similarity_model.onnx and detection_model.onnx) in the models/ directory.

Dependencies

•	Python 3.6 or higher
•	numpy
•	opencv-python
•	Pillow
•	onnxruntime

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contact

•	Author: Miao HanCheng
•	Email: [e-mail]