Logo image
結合圖型邊界地圖資訊的距離正規化水平集演化演算法-用於全自動的指骨輪廓圈選研究
Thesis

結合圖型邊界地圖資訊的距離正規化水平集演化演算法-用於全自動的指骨輪廓圈選研究

吳俊緯
Masters, 國立清華大學, 電機工程學系
2013

Abstract

影像切割 水平集演化 影像邊界 邊緣檢測函數 自動化 image segmentation level set evolution image edges edge indicator function automation
Abstract Distance Regularized Level Set Evolution, which is abbreviated to DRLSE, is an algorithm for image segmentation. In this thesis, we aim to improve the traditional DRLSE algorithm using edge indicator function, and develop a standard process using the proposed algorithm to separate phalanx from its background automatically. The key factor in the performance of DRLSE algorithm is the initial level contour. The closer the initial level contour and the outline of the target, the better result we get. The traditional DRLSE algorithm sets the initial level contour manually, which can be a tedious and inefficient work. In order to achieve better result, we propose a method using the predefined edge indicator function and several of image processing techniques to set the initial level contour automatically before traditional DRLSE evolution. Besides the general image segmentation problem, this study focuses on the problem of phalanx segmentation in particular. We proposed a standard process to separate phalanx from its background automatically by using the proposed modified DRLSE algorithm. The parameters setting of the algorithm are simplified, without the need for manual setting. The test images include some general images and a set of phalanx images. The result images, the execution time, the number of DRLSE iterations, and some measure of errors such as ME, RFAE, MHD, EMM, NU and Mean Errors are shown as experimental results. According to the experimental results, the proposed algorithm has three advantages over the traditional algorithm. First, the proposed algorithm sets the initial level contour automatically. It's more efficient and more convenient than the traditional one. Second, the whole computation time and the number of DRLSE iterations can be reduced drastically, which can avoid the waste of computing resources effectively. Third, the experimental results of the automated method for phalanx images have proved that the proposed algorithm works better. It can really help to enhance the effect of the DRLSE operation and reduces the error rates.

Metrics

1 Record Views

Details

Logo image