Center affine filter based adaptive image despeckling method with preserving details

Abstract

Speckle noises inherently exist in radar, sonar, ultrasound and optical coherent tomography images. Such noises greatly degrade image quality, and thus make the tasks such as object detection and image registration difficult. With the Center Affine Filter (CAF), this paper proposes a new adaptive method for suppressing speckle noises. The proposed method adopts both local arithmetic mean and local geometric mean around an image pixel to compute the dispersion parameter of the CAF. The metrics of arithmetic and geometric mean not only intimately relate to the probability density function of speckle noise, but also reflect the image homogeneity; therefore, the designed CAF is capable of adaptively suppressing the speckle noises while maintaining the useful details in an image. Then, to speed up the proposed method, the technique of the integral image is used to get the local arithmetic and local geometric mean. The proposed method is experimentally tested on both synthetic and real images, which shows a prospective performance over Lee filter, SRAD filter, and the state-of-the-art BNLM filter.