Bilateral filtering is a good alternative to conventional linear filtering, and finds several applications in image processing. Bilateral filters are nonlinear and have the property that they can smooth images without distorting the edge information. The edge-preserving property arises due to the fact that the filters not only take into account the spatial separation between pixels but also the differences in the corresponding amplitudes. The popular version of the bilateral filter employs a Gaussian kernel. In this project, we propose to construct a class of generalized bilateral filters employing kernels from the centered B-spline family. We shall reformulate the 2-D nonlinear bilateral filter as a linear filter in an augmented 3-D space. Spline kernels are perfectly suitable because they have compact support and also enable the development of fast recursive filtering algorithms. In the limiting case, as the order of the centered B-spline tends to infinity, the splines tend to a Gaussian. Therefore, our generalized formulation includes the bilateral Gaussian filter as a special case.
In the course of the project, the student will be required to develop new algorithms and implement them as plugins for ImageJ.
The target applications are in biological and medical image segmentation, and rendering artistic effects for photography applications.