Laplacian Pyramid Decomposition-Type Method for Resolution Enhancement of Ultrasound Images
Résumé
In biomedical ultrasound imaging, there is a great need of high-resolved images for diagnosis or therapeutic purpose. This paper presents a resolution enhancement algorithm based both on Laplacian pyramid decomposition and on autoregressive (AR) model prediction. The idea is to use the Laplacian pyramid decomposition to estimate the high frequency (HF) image. Classically, after a bicubic-spline interpolation and a correction by an empirical control function, this HF image prediction is added to the bicubic-spline interpolated low frequency (LF) image. The resulting image is improved but the bicubic-spline interpolations tend to smooth the speckle. As a consequence, an empirical correction function based on the original image histogram has to be added to the HF image prediction. To face these issues, we propose an alternative to the bicubic-spline interpolation using an AR model instead. The resulting image is enhanced, and the empirical control function is not needed any more. Both methods are compared on synthetic images with different noise levels and distances to resolve. The resolution improvement was quantified in each case using a resolution criterion and the PSNR through the Monte-Carlo method. Then, the two methods are applied on an in vivo 20 MHz ultrasound image and the effectiveness of the proposed algorithm is shown.