Speckle noise model and its optional removal in ultrasound 2D B-scan images

Summary form only given. Corruption of images by speckle noise reduces image resolution and changes the apparent reflectivity of tissues, resulting in improper tissue characterization. The authors propose a model for speckle as multiplicative noise and develop an adaptive algorithm for reducing it. The gray values of the pixels in a window are taken that are within a 2 sigma range of mean, and a median filter is applied. This algorithm adaptively changes the number of pixels taken to calculate median values, depending on local statistical properties. The filter has been applied to 25 normal and 25 abnormal renal scans. A discriminant classifier was applied following the provision of a set of training samples. It was observed that after optimum filtering, the tissue parameters for normal and abnormal renal scans could be well separated with improved confidence of identifying abnormal scans.<>