一個關於臨界值法和橢圓偵測法以及它們應用在影像切割的研究

Abstract

影像切割是一些影像處理程序之前的一個重要的工作。臨界值法和形狀偵 測法是作影像切割最常見技巧中的兩個方法。臨界值法主要可分為最佳化 臨界值法及基於特性分析的臨界值法。最佳化臨界值法存在以下的缺點： 演算法的計算時間成冪次方倍成長以及臨界點個數需事先給定。基於特性 分析的臨界值法則需指定平滑參數來產生臨界點。為了解決以上缺點，我 們提出了一些新的臨界值法。形狀辨識是從事影像切割另一個常用的方法 。過去提出來辨識橢圓的方法仍然有一些限制或需要很多的計算時間，因 此我們提出一個新的橢圓偵測法。為了顯示以上所提方法的可行性，我們 將之應用於醫學影像的切割。實驗結果顯示我們提出的臨界值法可應用於 肺部的切割，而橢圓偵測法則可切割出肺癌。 Image segmentation is an important task preliminary to some image processes. Thresholding and shape detection are two types of the most common approaches. There are two major thresholding approaches: optimal and property-based. For optimal thresholding methods, they suffer the disadvantages that the computing time grows exponentially with the number of thresholds, and the threshold number should be given in advance. For property-based thresholding methods, some of them output a series of thresholds based on specific values of the smooth parameters, but they do not provide a way to estimate these values. In this dissertation's study, three methods are proposed to solve individual problems. Shape detection is another popular approach to performing segmentation. The methods presented to detect ellipses still have some limitations or need a lot of computing time. For this reason, a new ellipse detector is proposed. To show the feasibility of the proposed methods, the segmentation of medical images is studied. Experimental results show that the proposed thresholding method is applicable for lung segmentation, and the proposed ellipse detector can extract the nodules.