GuasKDTreeFiltering 0.1.1

Spatial and Bilateral filtering using Gaussian KDTree data structures

GuasKDTreeFiltering

A high-performance Python library for efficient spatial and bilateral filtering using Gaussian KDTree data structures. This implementation is based on the research paper "Gaussian KD-Trees for Fast High-Dimensional Filtering" from Stanford Graphics Lab.

🚀 Features

• Spatial Filtering: Efficient spatial smoothing using KDTree and Gaussian kernels
• Bilateral Filtering: Edge-preserving smoothing for high-quality image processing
• High-Dimensional Support: Optimized for multidimensional data processing
• Fast Performance: Leverages KD-tree data structures for efficient neighborhood queries
• Easy Integration: Simple API for quick integration into existing workflows

📦 Installation

From PyPI (Recommended)

pip install GuasKDTreeFiltering

From Source

git clone https://github.com/enugalamanideepreddy/GuasianKDTreeFiltering.git
cd GuasianKDTreeFiltering
pip install -e .

🎯 Quick Start

Basic Usage

import numpy as np
from GuasKD import BilateralFiltering

# Create a bilateral filter instance
bf = BilateralFiltering(
    sigma_spatial=1.0,    # Spatial sigma
    sigma_color=0.1,      # Color sigma
    sigma_b=0.05,         # Blur sigma
    samples=32,           # Number of samples
    leaf_size=10          # KDTree leaf size
)

# Create test image (H, W, C format)
image = np.random.rand(100, 100, 3).astype(np.float32)

# Apply bilateral filtering
filtered_image = bf(image, image, mode='Bilateral')

Advanced Usage

from GuasKD import BilateralFiltering
import numpy as np

# Create filter with custom parameters
bf = BilateralFiltering(
    sigma_spatial=2.0,    # Larger spatial smoothing
    sigma_color=0.2,      # More color smoothing
    sigma_b=0.1,          # More blur
    samples=64,           # More samples for better quality
    leaf_size=8,          # Smaller leaves for precision
    fallback_mode='nearest'  # Fallback for edge cases
)

# Process different image types
image = np.random.rand(64, 64, 3).astype(np.float32)

# Bilateral filtering (preserves edges)
bilateral_result = bf(image, image, mode='Bilateral')

# Spatial filtering (smooths everything)
spatial_result = bf(image, image, mode='Spatial')

📚 API Reference

BilateralFiltering Class

Constructor Parameters

Parameter	Type	Default	Description
sigma_spatial	float	1.0	Spatial smoothing parameter
sigma_color	float	0.1	Color/intensity smoothing parameter
sigma_b	float	0.0	Blur parameter in KDTrees High Dimensional Filtering
samples	int	32	Number of samples for filtering
leaf_size	int	10	KDTree leaf node size
fallback_mode	str	'nearest'	Fallback strategy for edge cases
min_samples	int	4	Minimum samples per query

Methods

__call__(image, values, mode)

Main filtering method.

Parameters:

• image (np.ndarray): Input image in (H, W, C) format
• values (np.ndarray): Values to filter (usually same as image)
• mode (str): Either 'Bilateral' or 'Spatial'

Returns:

• np.ndarray: Filtered image with same shape as input

🔬 Research Background

This implementation is based on the research paper:

"Gaussian KD-Trees for Fast High-Dimensional Filtering"

• Authors: Stanford Graphics Lab
• Paper: https://graphics.stanford.edu/papers/gkdtrees/gkdtrees.pdf

The library provides an efficient implementation of the Gaussian KDTree approach for high-dimensional filtering operations, making it suitable for:

• Image denoising
• Edge-preserving smoothing
• High-dimensional data processing
• Real-time filtering applications

📄 License

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

🙏 Acknowledgments

• Stanford Graphics Lab for the original research
• The open-source community for inspiration and feedback

📞 Support

If you encounter any issues or have questions:

• Issues: GitHub Issues
• Documentation: Check the docstrings and examples above
• Research: Refer to the original paper for theoretical background

---

Made with ❤️ for the computer vision and image processing community