Adaptive Bilateral Texture Filter for Image Smoothing.

Area of Science:

• Image Processing
• Computer Vision

Background:

• Texture filtering aims to smooth image regions while preserving structural details.
• Current methods struggle to balance smoothing strong gradients and maintaining weak structures.

Purpose of the Study:

• To develop a scale-adaptive texture filtering algorithm that effectively addresses the challenge of smoothing strong gradients while preserving weak structures.
• To improve upon existing texture filtering techniques in terms of edge preservation and visual quality.

Main Methods:

• A four-directional detection method using gradient information for structure measurement.
• Adaptive spatial kernel scaling for each pixel based on local structure information.
• Utilizing the Fourier approximation of the range kernel to reduce computational complexity.

Main Results:

• The proposed algorithm successfully eliminates textures in textural regions by using larger spatial kernels.
• It effectively preserves edges and weak structures by employing smaller spatial kernels on structural pixels.
• Subjective and objective analyses demonstrate superior performance compared to previous methods.

Conclusions:

• The scale-adaptive texture filtering algorithm offers an effective solution for smoothing strong gradients while preserving weak structures.
• The method achieves high structure similarity and visual perception quality, outperforming existing techniques.
• Computational complexity is reduced without compromising filtering quality.