Image synthesis based on manifold learning

Abstract

A new methodology for image synthesis based on manifold learning is proposed. We employ a local analysis of the observations in a low-dimensional space computed by Locally Linear Embedding, and then we synthesize unknown images solving an inverse problem, which normally is ill-posed. We use some regularization procedures in order to prevent unstable solutions. Moreover, the Least Squares-Support Vector Regression (LS-SVR) method is used to estimate new samples in the embedding space. Furthermore, we also present a new methodology for multiple parameter choice in LS-SVR based on Generalized Cross-Validation. Our methodology is compared to a high-dimensional data interpolation method, and a similar approach that uses low-dimensional space representations to improve the input data analysis. We test the synthesis algorithm on databases that allow us to confirm visually the quality of the results. According to the experiments our method presents the lowest average relative errors with stable synthesis results. © 2011 Springer-Verlag.