| dc.contributor | 統計系 | |
| dc.creator (作者) | 吳漢銘 | |
| dc.creator (作者) | Wu, Han-Ming | |
| dc.creator (作者) | 李昱儀 | |
| dc.date (日期) | 2019-03 | |
| dc.date.accessioned | 2022-04-12 | - |
| dc.date.available | 2022-04-12 | - |
| dc.date.issued (上傳時間) | 2022-04-12 | - |
| dc.identifier.uri (URI) | http://nccur.lib.nccu.edu.tw/handle/140.119/139846 | - |
| dc.description.abstract (摘要) | 影像切割技術在影像樣態辨識過程中是一個很基礎且重要的步驟，其應用的層面甚廣，例如醫學影像腫瘤診斷或衛星圖像定位等等，因此如何提昇影像切割的正確率，一直以來都是很熱門的研究議題。當在進行影像切割時，通常是面對巨量且高維度的資料處理，藉由維度縮減法來改善效能及提高正確率是其中一種解決方案。等軸距切片逆回歸法（ISOSIR）是一個有效的非線性維度縮減法，它可以找出高維度資料中所隱藏的特徵並呈現資料的幾何結構於低維度空間。ISOSIR主要的特點是使用K均值分群法（KMS）將資料的等軸距距離矩陣作分群，然後結合切片逆回歸法來應用在分類問題上，相較於傳統方法，例如等距映射圖法（ISOMAP）或主成份分析（PCA），有最佳的表現。影像切割問題即是應用分群或分類方法於影像特徵資料上的結果。雖然ISOSIR演算法中，運用KMS的效果良好，然而，KMS這種分群方法在重要且細微的影像資料特徵分群上易有偏差。因此，本研究將進一步採用模糊C均值分群法（FCM）來做爲切片逆回歸法分群的依據。同時，針對影像產生的大量特徵資料，我們提出地標等軸距模糊切片逆回歸法（L-ISOFSIR）來改善計算的效能。我們考量了三種不同影像特徵的擷取方法，應用所提的新方法於二組模擬影像及一組實際醫學影像的切割問題，並與FCM及模糊切片逆回歸法相比較。實例結果顯示本研究所提出的新方法L-ISOFSIR可以顯著地增進影像切割的正確率，是一個有效率的電腦自動影像切割方法。 | |
| dc.description.abstract (摘要) | Image segmentation is an essential and crucial step in the pattern recognition processes. It has been applied to many fields such as the medical image segmentation for tumor diagnosis and the satellite image positioning. The approaches to improve the accuracy of image segmentation have become an active research topic. When conducting the image segmentation, the features extracted from images usually consist of a large amount of high-dimensional data. Dimensionality reduction (DR) is one of the solutions to such data and is employed to improve the efficiency and the accuracy of the segmentation. Among many DR methods, the isometric sliced inverse regression (ISOSIR) is an effective non-linear method that can be used to discover the embedded features of the high-dimensional data with their geometric structure is presented in the low-dimensional subspace. ISOSIR uses K-means (KMS) to cluster the isometric distance matrix of the input data and then applies SIR to the classification problems. It has been proved to perform better than some traditional DR methods such as ISOMAP or PCA. Although KMS can achieve good performance, it tends to be biased for clustering the image features. Consequently, we use the fuzzy c-means clustering (FCM) as an alternative for SIR in this study. For the large amount of features generated from the images, we are motivated to propose the landmark isometric fuzzy sliced inverse regression (L-ISOFSIR) to improve the computational efficiency of the image segmentation. Two sets of simulated images and one real medical image are used to evaluate the proposed method based on three feature domains. Comparisons with results obtained via FCM and the fuzzy sliced inverse regression (FSIR) are also reported. The experimental results show that L-ISOFSIR improves the accuracy of image segmentation significantly which is an efficient computer-aided automatic image segmentation tool. | |
| dc.format.extent | 7736907 bytes | - |
| dc.format.mimetype | application/pdf | - |
| dc.relation (關聯) | 中國統計學報, Vol.57, No.1, pp.43-70 | |
| dc.subject (關鍵詞) | 模糊C均值法;等軸距切片逆回歸法;非線性維度縮減法;影像切割 | |
| dc.subject (關鍵詞) | Fuzzy c-means clustering;Image segmentation;ISOMAP;Nonlinear dimensionality reduction;Sliced inverse regression | |
| dc.title (題名) | 地標等軸距模糊切片逆回歸法及其於醫學影像切割的應用 | |
| dc.title (題名) | Landmark Isometric Fuzzy Sliced Inverse Regression with Application to Medical Image Segmentation | |
| dc.type (資料類型) | article | |
