Arbitrarily-Oriented Anisotropic 3D Gaussian Filtering Computed with 1D Convolutions without Interpolation
| Authors | |
|---|---|
| Year of publication | 2008 |
| Type | Article in Proceedings |
| Conference | Proceedings of 8th WSEAS International Conference on Signal Processing, Computational Geometry and Artificial Vision |
| MU Faculty or unit | |
| Citation | |
| Field | Informatics |
| Keywords | Gaussian separability spatial image filtering |
| Description | The paper presents a procedure that allows for good approximation of anisotropic arbitrarily-oriented arbitrarily-dimensional Gaussian filter. Even though the method is generally $n$D, it is demonstrated and discussed in 3D. It, essentially, substitutes a given 3D Gaussian by six 1D oriented recursive convolutions. Since many such orientations may be available, a set of fast constraints is presented to narrow the number of them and to select the best ones. Firstly, only integer elements are allowed in the directional vector of each orientation not to break translation-invariance of the method. Secondly, validations due to requirements on separability of the approximated filter are applied. Finally, the best directions are selected in the convolution test. We compare the method to a similar recent approach of [Lampert and Wirjadi, 2007] in 3D. A possibility how to improve the performance of bank filtering by using this method is outlined. |
| Related projects: |