Directional image filter respectively adjusted to edge and flat regions

Abstract

The authors propose a directional filter for eliminating noise from MRI images. Kido et al. developed a directional filter capable of detecting the minimal change direction in each image point, and of performing one‐dimensional (1D) nonlinear smoothing limited by that minimal change direction. This filter was effective in application to MRI images. In the current study, the authors improve this directional filter and develop an image filter for directional nonlinear smoothing that separates processing of edge and flat regions in an image. For the edge regions, this filter focuses on edge structure obscured by noise. It involves obtaining the group minimal change direction, taking into account the image structure in the vicinity of the current image element, and performing 1D nonlinear smoothing. This processing allows smoothing that produces neatly aligned edge directions for the adjoining image elements, even in sections where the image edge structure is obscured by noise. In addition, for the flat regions, the smoothing effect is enhanced by performing 1D nonlinear smoothing over a distance longer than that for the edge regions. Application to MRI images produces good images with neatly aligned edge directions, and enhanced smoothing effect in flat regions where excessive smoothing of the edge regions has been suppressed. ©1999 Scripta Technica, Syst Comp Jpn, 30(8): 72–78, 1999

An Adaptable $k$-Nearest Neighbors Algorithm for MMSE Image Interpolation

“…Though texture results in high peak signal to noise ratios (PSNR), to be presented in Section V, the human visual system (HVS) focuses on edges [48]. Fortunately, research into edgeoriented image filtering has been well-studied [49], [50]. In our framework, we can agglomerate a bank of edge-oriented filters that do "well-enough" when the "best" filter through -NN is unavailable, effectively reducing the implementation to a specialized version of [9] with an added MRF improvement (see the next section, Section IV) through [51].…”