Abstract:

To extend the application of semisupervised learning in image segmentation, an image segmentation method based on the manifold is proposed. This approach interprets the classification problem as a problem of interpolating a function on a manifold. Some coefficients are adjusted to provide the optimal fit to the data. The algorithm makes use of sparse adjacency matrix, which makes solving eigenvector problems for big matrix possible. However, it takes long time to construct the sparse adjacency matrix for image segmentation. To reduce computational complexity, an approach is proposed based on the Nystrm method, a numerical solution of eigenfunction problems. Experimental results of synthetic texture images segmentation indicate that the proposed method achieves good quality and using Nystrm method improves the computational efficiency to a great degree.