視覺模型與區域競爭法在超音波影像上的應用

Abstract

本篇論文的目的是希望透過影像處理中的區域競爭法來進一步結合超音波 影像關於紋路與雜訊的兩項特質. 針對紋路, 模仿人類視覺模型所產生的 視距圖(distance map)可以提供邊界的資訊. 超音波影像的雜訊則可根據 其形成原理而導出Rayleigh 分佈的統計模型. 區域競爭法引進概似化檢 定(likelihood ratio test), 令邊界上的點沿著最陡的下降(steepest descent) 的方向移動, 以達到最小化能量函數的目的. 我們進一步對於 邊界不甚明顯的區域進行修正, 以得到一個完整的區域. Ultrasound images are difficult for analysis due to their complex textures and speckle noises. In this study, it is aimed to combine these two characters with a region-based technique named ``region competition" to analyze ultrasound images. For textures, a vision model is applied to generate a distance map which provides peak points for edge information. The Speckle noises are modeled by the Rayleigh distribution or its transformation based on the statistical properties of speckle noises. Region competition employs the likelihood ratio tests to move pixels on the boundary along the steepest descent direction in order to minimize the total energy function. These three techniques are integrated into a new approach for the segmentation of ultrasound images. Special treatments of weak edges are suggested by adjusting the weighting parameter adaptively. Simulation and emperical studies are performed and reported to investigate the feasibility of this new approach.