SAR segmentation using anisotropic diffusion

Abstract

Speckle effects are commonly observed in a synthetic aperture radar (SAR) image, which applies in many land applications, such as agriculture or forestry. However, although the human eye is able to derive the meaningful information from the SAR image, automatic processing is very difficult to distinguish the objects in the images because of the noise effect present in the image. This paper presents a segmentation method of the SAR images, which uses an anisotropic diffusion algorithm with an inherence region-growing algorithm. Anisotropic diffusion is modified in logarithmic domain to remove the speckle noise and to identify the homogeneous regions in ERS and JERS satellite data. The modified diffusion method is formulated with a tensor technique, varying with both the location of a pixel in the image and the orientation of the local image geometry vicinity the pixels. In our algorithm, SAR images are transformed into logarithmic domain and using the modified anisotropic diffusion to grow the homogeneous regions in the noise environment; consequently, the segmented image in the logarithm domain is converted to the intensity domain by an exponential function. On the other hand, our diffusion algorithm can stop itself by a standard deviation to mean criteria. Simulation results on the satellite images are provided to indicate the effectiveness of the proposed method when used to remove the effects of speckle noise almost as well as that of the additive noise. The evaluation of the proposed method employs the theoretical gain in equivalent numbers of looks (ENL). © 2005 SICE.