Regularized Speckle Reducing Anisotropic Diffusion for Feature Characterization
Yongjian Yu, Joseph Yadegar · 2006
For tissue characterization in medical ultrasound imagery or terrain characterization in synthetic aperture radar imagery, it is necessary to preprocess imagery to reduce granular, texture-alike noise called speckle. This preprocessing is difficult when it is needed to preserve delicate image details that are buried in speckle. Speckle reducing anisotropic diffusion (SAR) is a partial differential equation-based method developed for this purpose. Toward its improved performance for point/linear features, we introduced a novel regulator called energy condensation integral and developed a regularized SRAD (reg-SRAD) via minimization. The reg-SRAD generates outputs with increased resolution for point and linear features while retaining the characteristics the SRAD-filtering speckle with regional features enhanced. The performance of the method has been illustrated using synthetic and real ultrasound data, and radar imagery as well.