Separation of specular and diffuse components using tensor voting in color images

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Specular reflection removal is challenging in highly textured, multicolored images due to non-uniform highlights and artifacts.
• Existing methods often struggle with discontinuities in surface colors, leading to imperfect results.

Purpose of the Study:

• To propose a novel, noniterative, and constraint-free method for detecting and removing specular reflections from single color images.
• To address the limitations of current specular reflection removal techniques, particularly in complex image scenarios.

Main Methods:

• A novel method based on tensor voting is introduced for highlight detection and removal.
• Tensors' saliency analysis is employed to differentiate diffuse and specular pixels, avoiding neighbor pixel color comparisons.
• The derived diffuse reflectance distribution is utilized to effectively remove specular components.

Main Results:

• The proposed tensor voting method successfully detects and removes highlight components.
• Quantitative and qualitative evaluations on textured, multicolor images demonstrate superior performance.
• The method outperforms existing state-of-the-art techniques in specular reflection removal.

Conclusions:

• The novel tensor voting approach provides an effective solution for specular reflection removal in challenging images.
• This method offers a significant advancement over traditional techniques, particularly for complex visual data.
• The constraint-free and noniterative nature makes it a practical tool for image processing applications.