Fast edge-preserving noise reduction for ultrasound images

Edge-preserving noise reduction is an essential operation for computer-aided ultrasound image processing and understanding. This paper describes a novel filter which is a two-dimensional (2-D) extension of the one-dimensional (1-D) Savitzky-Golay filter. The new filter, referred to as the 2-D weighted Savitzky-Golay filter, is based on the least squares fitting of a polynomial function to image intensities. The performance of the proposed filter has been compared with that of the commonly used median filter in reducing speckle noise on a synthetic image and an ultrasound thyroid image. Experimental results indicate that on these particular examples, the new filter can achieve at least the same level of noise reduction and edge preservation as that of the median filter, but with far less computation time. Since its complexity scales linearly with the problem size, the new filter is suitable for filtering problems with large windows. In addition, its performance also shows to be less sensitive to the size of the filtering window compared to the median filter.