An Edge-Preserving Filter for Imagery Corrupted with Multiplicative Noise
Heather North, Qianqian Wu · Photogrammetric Engineering & Remote Sensing · 2001
In the segmentation of natural imagery, differentiation at feature boundaries is of crucial importance. The high-amplitude, multiplicative speckle noise present in synthetic aperture radar (SAR) data demands a high level of filtering, yet this noise must be removed without destroying the critical feature boundary information. We previously designed the minimum coefficient of variation (MCV) filter to meet the twin demands of noise removal and edge preservation in SAR imagery. MCV-filtered images exhibit clear feature boundaries, but the filter's strong edge-preserving nature also introduces step edge artifacts in areas of intensity gradient and texture. We present the modified MCV filter (MMCV) which is able to significantly reduce the occurrence of filtering artifacts, while retaining an edge-preserving character. The MMCV filter is compared to existing filters by operating them on SAR imagery and deriving edge maps from the filtered imagery. Though the MMCV-filtered image is not the most visually pleasing, the linework derived from it is the most useful in terms of clean, continuous feature boundaries.