Implementation of Asymmetric Kernel Median Filtering for Real-Time Ultrasound Imaging

Abstract

Ultrasound images contains speckles that could be considered as salt-and-pepper noise. Median filtering can efficiently remove the noise without blurring the edges. However, median filtering is computationally intensive and not suitable for real-time imaging. In this paper, median filtering on the intermediate ultrasound data using asymmetric median kernel sizes is proposed. To reduce the computational time, the median selection networks with compare-and-swap stages are applied instead of reducing the redundancy of the median calculation among the adjacent image pixels. Moreover, the modification of the compare-and-swap stages to suit for the GPU is also proposed. The computational times using CPU and GPU programs are compared and CNRs of the ultrasound images are also reported.