Handling noise in single image defocus map estimation by using directional filters.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Defocus map estimation is crucial for various imaging applications.
• Existing methods are highly sensitive to image noise, leading to significant performance degradation.
• Standard denoising techniques can alter edge profiles, compromising defocus estimation accuracy.

Purpose of the Study:

• To develop a robust method for defocus map estimation from noisy images.
• To overcome the limitations of existing noise-sensitive and edge-altering approaches.
• To improve the accuracy and reliability of defocus map estimation in the presence of noise.

Main Methods:

• A novel approach utilizing directional low-pass filters to reduce noise while preserving edges.
• Applying a series of directional filters at multiple orientations to the input image.
• Estimating blur amount along edges orthogonal to filter directions.
• Employing an edge-aware interpolation method to propagate blur information for a complete defocus map.

Main Results:

• Directional filtering significantly reduces image noise.
• Edges orthogonal to the filter direction are well-preserved, enabling accurate blur estimation.
• The proposed method demonstrates superior performance compared to state-of-the-art techniques on both synthetic and real-world noisy data.

Conclusions:

• The proposed directional filtering and edge-aware interpolation method offers a robust solution for defocus map estimation in noisy conditions.
• This technique effectively balances noise reduction and edge preservation, outperforming existing methods.
• The findings have implications for improving image quality and analysis in challenging imaging environments.