Approximation of an arbitrary filter and its recursive implementation

摘要

Filtering is one of the important techniques in computer vision. It has been widely used in edge detection, image restoration, range image segmentation, etc. However, the efficient implementation of an arbitrary filter has been a challenging problem until now. In this paper, a novel method is proposed to implement an arbitrary filter. Firstly, an efficient recursive structure is proposed to implement any (polynomial) × (exponential)-type (PET) filter. The computational complexity and structure are independent of its filter mask size or its bandwidth. Secondly, a new method is proposed—Lagurre spectrum decomposition method—to obtain the PET approximation of any filters. As an example, the above method is applied to the approximation and implementation of Gaussian filters and experiments have shown that a perfect approximation can be obtained with only third-order Lagurre bases, and therefore only a fourth-order recursive filter is needed to implement Gaussian filters. Finally, the comparison of the present method with the known ones shows that (1) Lagurre polynomial bases are orthogonal with each other, so the filter approximation is simple, (2) the bases are complete and the completeness guarantees the approximation error can be reduced to zero, (3) the method can be used to design both Gaussian and any other filters.