PROJECT TITLE :

A Continuous Random Walk Model With Explicit Coherence Regularization for Image Segmentation

ABSTRACT:

Images can be segmented using the random walk algorithm, which is a popular and efficient method (ROIs). When applied to large photos, the random walk approach requires solving a massive sparse linear system. For one thing, it's quite sensitive to seed distribution, which makes it difficult for users to get accurate segmentation results. A continuous random walk model with explicit coherence regularisation (CRWCR) for the extracted ROI is proposed in this research, which helps to reduce the seed sensitivity, thereby reducing user interactions. Finally, a highly efficient technique for solving the model will be created, which removes the challenge of tackling large linear systems. First, we run a one-dimensional random walk sweep using user-provided seeds, followed by a Peaceman-Rachford alternating-direction algorithm (PRA) for additional correction. Using a small number of one-dimensional random walk problems, we can quickly get a good idea of ROI in the first stage. Once this initial prediction has been refined, the second stage is applied, which is likewise very efficient because it works well with GPU processing, and 10 iterations are usually adequate for convergence. The proposed model and the efficiency of the two-stage procedure are tested numerically.