STAR: A Structure and Texture Aware Retinex Model.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Retinex theory decomposes images into illumination and reflectance components.
• Local image derivatives are key to traditional Retinex-based methods.
• Existing methods face challenges in accurately separating these components.

Purpose of the Study:

• To propose a novel Structure and Texture Aware Retinex (STAR) model.
• To enhance image decomposition by utilizing exponentiated local derivatives.
• To improve low-light image enhancement and color correction.

Main Methods:

• Generating structure and texture maps using exponentiated local derivatives (exponent γ).
• Designing exponential filters for local derivatives.
• Employing structure and texture maps to regularize Retinex decomposition.
• Solving the STAR model via an alternating optimization algorithm with vectorized least squares regression.

Main Results:

• The STAR model effectively extracts accurate structure and texture maps.
• Demonstrated superior quantitative and qualitative performance compared to existing methods.
• Achieved improvements in illumination/reflectance decomposition, low-light enhancement, and color correction.

Conclusions:

• The proposed STAR model offers a significant advancement in single-image decomposition.
• STAR's use of exponentiated derivatives provides better regularization for Retinex.
• The method shows broad applicability in image enhancement and correction tasks.