1 INTRODUCTION The presence of speckle noise in ultrasound im- ages is common and strongly interferes in detecting and analyzing the anatomical structures in these im- ages. The main type of noise found in ultrasound images appears due to the overlapping and deviating of the sound waves emitted and received by the ul- trasound sensors. Hence, the study of techniques to reduce this type of noise is important as it allows the understanding of the corrupted images more effi- ciently. Several computational image processing tech- niques have been proposed to remove image noise, including techniques based on Gaussian filters [1, 2], differential equations [3, 4, 5, 6] and multi- resolution processing [7]. Usually, the speckle noise is more complex to be removed than Gaussian noise; for this reason, it is common to represent the speckle noise as additive noise. Thus, there are various works that try to remove speckle noise based on ap- proaches originally developed to smooth images af- fected by additive noise. Here, a novel selective smoothing method for im- ages affected by speckle noise is proposed. The method uses the radiation intensity of the pixels of the image to be smoothed, and tries to find the pixels that belong to potential contours in order to perform a selective image smoothing using average filters with different intensities. This paper is organized as follows: the proposed smoothing method is described in the next section, the experimental results and their discussion are pre- sented in section 3, followed by the final considera- tions in the last section. 2 PROPOSED METHOD The new denoising method is based on selective average filtering of the input image using average filters, and defines the intensity addressed by each filter according to the radiation accumulated by each image pixel from its neighbor pixels. The method starts the noise removal process by weakly smooth the input image using an average fil- ter with a 3x3 convolution window, which avoids the loss of important details presented. In the next step, the radiation intensity received by each image pixel is computed by summing all interferences caused by the neighbor pixels, using the following equation: , 4 2 1 2 ∑ =       −         = w i m i i T r I R π (1) where i I is the pixel i of image I and m T is a regu- lating term. The radiation intensity is computed in a circular region of radius r , which starts with value equal 1 (one) and grows until 2 w pixels, where w is the width of the input image. In this computation, only are considered the radiations associated to the neighbor pixels at the same line, column and diago- A selective denoising method to remove speckle noise Alex F. de Araujo & João Manuel R. S. Tavares Instituto de Engenharia Mecânica e Gestão Industrial, Faculdade de Engenharia, Universidade do Porto, R. Dr Roberto Frias s/n, 4200-465 - Porto, PORTUGAL fa.alex@gmail.com, tavares@fe.up.pt Christos E. Constantinou Department of Urology, School of Medicine Stanford University, Stanford, CA, USA constantinou@stanfordalumni.org ABSTRACT: Speckle noise strongly affects the images acquired by ultrasound scans and reduces the effi- ciency of computational methods to extract and understand the features presented in the images. Trying to overcome this problem, a selective denoising method is proposed based on average filtering and on the radia- tion intensity associated to the input image pixels. Experimental tests were done using a set of simulated ul- trasound images and a real ultrasound image sequence, and their statistical analysis confirmed that the pro- posed method has good ability to reduce speckle noise in ultrasound images.