Multi-window Local Variance Feature Detection Based Ultrasound Image Sharpening

Ultrasound images suffer speckle noises and several artifacts that degrade image contrast and cause difficulties in observing borders and features. In this paper, we propose an ultrasound image sharpening algorithm that actively enhances edges and borders in ultrasound images. The sharpening is done by amplifying the high-frequency features in the image. However, ultrasound images contain speckle noises which are also high frequency components. To avoid the undesired sharpening of speckle noises, a feature mask is needed to adaptively adjust the enhancing strength and bypass homogeneous speckle regions. The feature mask is provided via a novel multi-window local variance (MLV) feature detection technique that utilizes the constant variance characteristic of Fisher-Tippett distribution in log-compressed speckle regions and could detect features robustly on ultrasound images. The proposed image sharpening algorithm is tested on simulated and clinical ultrasound B-mode images and the results show improvements in objective metrics and subjective evaluations. In addition, we demonstrate that the proposed image sharpening technique can be cascaded with other enhancement algorithms and compensate for the blur caused by speckle reduction algorithms.