ROBUST WEIGHTED REGRESSION FOR ULTRASOUND IMAGE SUPER-RESOLUTION

Reconstructing a high resolution image from a low resolution one is an under-determined problem without unique solutions. Many different approaches have been proposed. In this paper we focus on images from a specific application domain the gray scale ultrasound images. Ultrasound images have unique structures, i.e., they contain large patches of the same color. We use the square neighborhood of a pixel to predict its corresponding pixels in the high resolution version through a regression model. The regression model is weighted due to heteroscedasticity observed in fitting regression models to ultrasound images. Our weighted regression approach have very good performance.