Please use this identifier to cite or link to this item:
標題: 基於支持向量機輔助之自我組織模糊網路做膚色影像分割
Skin Color Image Segmentation by Support Vector Machine-aided Self Organizing Fuzzy Network
作者: 邱士軒
Chiu, Shih-Hsuan
關鍵字: Support Vector Machine;支持向量機;Neural Network;Fuzzy Network;Cluster;Skin Color Image Segmentation;類神經網路;模糊網路;分群;膚色影像;影像切割;分類器;直方圖
出版社: 電機工程學系
本論文提出基於支持向量機輔助之自我組織模糊網路(SVM-SOFN)做膚色影像分割。SVM-SOFN是一個結合模糊分群(Fuzzy Clustering)與支持向量機(SVM)的模糊系統,此模糊系統的前件部為對輸入資料作模糊分群而得,再使用支持向量機來做後件部的調整,使得整個系統能夠得到較佳的性能。在SVM-SOFN中,有兩種型式的後件部可供選擇,首先是模糊單值型式,稱之為SVM-SSOFN,另一個則是TSK 型式,稱之為SVM-TSOFN。在特徵值的方面,我們使用了HSV色彩模型中的色調和濃度二維彩色空間來表示每一彩色像素。為增加以直方圖統計表示色彩的精確度,我們使用非均勻的HS空間切割方式。訓練SVM-SOFN所用的直方圖統計資料乃是由不同環境所拍的影像來獲得,以便增加方法的強健性。為測試所提方法的好壞,我們將其應用在膚色切割上,並比較了其他切割方法,其中包含了基於直方圖統計方式的膚色分類器、高斯混合分類器、自我建構類神經模糊推論網路和支持向量機四種方法。在這些分割的方法裡,我們發現使用SVM-TSOFN方法是最好的而且可得到不錯的分割結果。

Skin color image segmentation by Support Vector Machine-aided Self Organizing Fuzzy Network (SVM-SOFN) is proposed in this thesis. SVM-SOFN is a fuzzy system constructed by the hybrid of fuzzy clustering and SVM. Antecedent part of SVM-SOFN is generated via fuzzy clustering on the input data, and then SVM is used to tune the consequent part parameters to make the network with better generalization performance. In SVM-SOFN, there are two types of consequent parts that can be used. One is SVM-aided singleton type SOFN, called SVM-SSOFN, and the other is SVM-aided TSK-type SOFN, called SVM-TSOFN. Each color pixel is represented by hue and saturation component of HSV color space. To represent a color by histogram as accurately as possible, non-uniform partition of HS space is used. Histogram information from images under different environments is used to train SVM-SOFN to make the method as robust as possible. To verify performance of the proposed method, experiments on skin color segmentation are performed. For comparison, other four color image segmentation methods, including Histogram-based Skin Classifier, Mixture of Gaussian Classifier, Self-cOnstructing Neural Fuzzy Inference Network, and Support Vector Machine, are applied to the same problem. From comparisons, we find that SVM-TSOFN achieves the best segmentation result.
Appears in Collections:電機工程學系所

Show full item record

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.