結合圖像分割與機率類神經網路於不同膚色的切割及其應用於手勢分類之研究

機構典藏 > College of Engineering > Graduate Institute & Department of Electrical Engineering > Thesis > Item 987654321/74705

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/74705

Title:	結合圖像分割與機率類神經網路於不同膚色的切割及其應用於手勢分類之研究
Other Titles:	Segmmentation of different skin colors by combining graph cuts with probability neural network, and its application to classification of hand gesture
Authors:	呂愷迪;Lu, Kai-Di
Contributors:	淡江大學電機工程學系碩士班黃志良;Hwang, Chih-Lyang
Keywords:	膚色切割;圖像分割;機率類神經網路;貝氏分類器;Skin color segmentation;Graph cuts;Probability neural network;Bayesian classifier
Date:	2011
Issue Date:	2011-12-28 19:21:55 (UTC+8)
Abstract:	近年來人機介面(human computer interaction)的技術日新月異，遙控器類的控制方法已漸漸的不符潮流，而偏向更人性化的控制，如體感遊戲機。其應用視覺系統辨識人的手勢或姿態以控制相關機器。然而利用手勢或人的姿態來控制機器之前，必須先從影像畫面中切割出正確的膚色目標，藉此才能辨識出正確的手勢或姿態，而如何能從影像中正確切割出皮膚顏色的目標即為本論文研究的重點。切割皮膚顏色之所以困難，是因為皮膚顏色範圍相當廣，不同的皮膚顏色由黑到白都有，且還會受到不同的燈光條件影響，放寬切割膚色的閥值又容易切割出日常生活中類似皮膚顏色的物品，所以如何設計出一種演算法則，能克服以上種種變因，一直為大家熱烈討論及研究的課題。　　本論文利用機率類神經網路(Probability Neural Network)和圖像分割法(Graph Cuts)設計出一種最佳切割閥值的系統，目的為切割複雜背景中不同的皮膚顏色。相較於以往利用固定閥值切割皮膚顏色，本論文提出以圖像分割法估算出的動態閥值較能克服不同的皮膚顏色以及不同的環境。因為圖像分割法是以輸入影像為基礎估算出的最佳化閥值，較能容納背景環境改變及皮膚顏色在不同燈光下等的變因，固定閥值則無法藉由外在環境改變而調整其閥值，比較局限於特定情況下的切割。但假如僅僅使用圖像分割法在如下情況可能會造成失敗：任何類似膚色的目標，只要屬於此閥值以內皆會被程式判定為膚色目標。　　為解決此問題，加入了機率類神經網路來對程式判定的可能皮膚目標做分類，進一步過濾掉類似皮膚顏色的非膚色目標。因此首先調查各個皮膚顏色在色彩空間的分布情況，再以機率類神經網路的前置訓練，以獲得具有分類能力的類神經網路。本論文將色彩空間內的所有色彩分為四類，分別為黑皮膚顏色、黃皮膚顏色、白皮膚顏色和非皮膚顏色。則可以對圖像分割法所獲得的結果進一步篩選，更提升膚色辨識的正確率和對複雜背景及不同燈光下的強健性，最後在將我們的方法應用於八種不同的手勢分類。 It is known that fixed thresholds mostly fail in two situations as they only search for a certain skin color range: (i) any skin-like object may be classified as skin if skin-like colors belong to fixed threshold range. (ii) any true skin for different races may be mistakenly classified as non-skin if that skin colors do not belong to fixed threshold range. In this paper, a dynamic threshold of different skin colors based on the input image is determined by the combination of graph cuts (GC) and probability neural network (PNN). The compared results among GC, PNN and GC+PNN are presented not only to verify the accurate segmentation of different skin colors but also to reduce the computation time as compared with only using the neural network for the classification of different skin colors and non-skin color. The experimental results for different lighting conditions also verify the usefulness of our method. Finally, the application to the classification of hand gestures in complex environment is presented to evaluate the effectiveness and efficiency of the proposed method.
Appears in Collections:	[Graduate Institute & Department of Electrical Engineering] Thesis

Files in This Item:

File	Size	Format
index.html	0Kb	HTML	206	View/Open

Loading...