Solving ill-posed image processing problems using data assimilation (Q625544)

In image processing most problems are ill-posed in the sense that it is not possible to provide a unique solution. This paper proposes a strategy to solve some of these problems by a method called ``data assimilation''. It aims to solve an evolution equation, describing the temporal dynamics, and an observation equation, linking the state vector and the observations. The paper is organized in six sections and three appendices. Section 1 introduces the concept of ill-posed problems, a brief description of some of the strategies commonly used to solve them and the method called data assimilation. Section 2 defines mathematically the concepts introduced in Section 1. Section 3 describes the variational data assimilation method known as the 4D-Var algorithm. Section 4 explains how data assimilation can be used to solve ill-posed problems by assimilating images in an appropriate evolution model. Section 5 describes how to compute optical flows in this framework. Section 6 provides scientific perspectives of this research. The Appendices provide technical details of the issues discussed above. While the treatment of the issues is formally rigorous, ``generalized functions'' are not used, which would be more appropriate from the point of view of mathematics. However, the paper is well written and will appeal to those working with spatial and temporal patterns in image processing and analysis.