Constrained pseudo-Brownian motion and its application to image enhancement

Area of Science:

• Computer Vision
• Image Processing
• Computational Neuroscience

Background:

• Brownian motion is a random process with applications in various fields, including color perception.
• Existing retinex algorithms based on Brownian motion paths, such as Marini and Rizzi's, offer advantages but suffer from high computational complexity.
• Multiscale retinex algorithms like McCann99 and Frankle and McCann's provide low computational complexity but can introduce artifacts.

Purpose of the Study:

• To develop an efficient algorithm for generating pseudo-Brownian paths for retinex implementations.
• To ensure statistical similarities between generated paths and true Brownian motion or random walks.
• To reduce computational complexity and artifacts in retinex-based image processing.

Main Methods:

• Proposed an efficient algorithm to generate pseudo-Brownian paths with guaranteed lower bounds on pixel visit counts.
• Ensured statistical similarities to random walk and Brownian motion.
• Implemented a retinex algorithm utilizing the generated pseudo-Brownian paths.

Main Results:

• The generated pseudo-Brownian paths exhibit statistical similarities to random walks and Brownian motion.
• The retinex implementation using the new algorithm produced fewer artifacts compared to McCann99 and Frankle and McCann.
• The proposed method required fewer pixel comparisons for comparable results, compensating for slightly increased computational complexity.

Conclusions:

• The developed algorithm offers an efficient and effective method for generating pseudo-Brownian paths for retinex applications.
• This approach enhances image processing by reducing artifacts and improving computational efficiency.
• The study demonstrates a promising alternative for retinex implementations in color perception and image analysis.