使用統計直方圖與類神經網路在彩色影像中偵測皮膚的比較

CCUR > College of Natural Science and Engineering > College of Engineering > College of Engineering > Chinese Culture University Hwa Kang Journal of Engineering > Item 987654321/21911

Please use this identifier to cite or link to this item: https://irlib.pccu.edu.tw/handle/987654321/21911

Title:	使用統計直方圖與類神經網路在彩色影像中偵測皮膚的比較
Authors:	逄霖生、李家崴、張明朗、翁志祁
Contributors:	工學院
Keywords:	RGB YCbCr Ratio 直方圖閥值 BPNN RGB Histogram Thresholding, BPNN
Date:	2009-01-01
Issue Date:	2012-03-28 14:20:53 (UTC+8)
Abstract:	本文以兩種模型在色彩空間中偵測皮膚並作分析比較。主要是利用RGB和CbCr各種色彩的比值(ratios)為主要特徵，比較皮膚和非皮膚兩者之間分佈關係以作為辨識的準則。在偵測的過程中我們先將取得的照片以人為的方式區分出皮膚與非皮膚兩個部分，這個區分的過程在本文稱知為全盤的已知。緊接著區分皮膚和非皮膚各色彩比值的資訊，再分別探討它們的關係。第一種模型以統計直方圖(histogram)來做分析比較，從中得知之間的顏色特性，並以此作為皮膚偵測特徵，以事先選擇的閥值(thresholding)作為偵測的標準，符合條件的像素點視為皮膚，反之為非皮膚。接者本文以倒傳遞類神經網路(Backpropagation Neural Network, BPNN)來偵測皮膚。先在特定的圖像中訓練神經網路，然後利用訓練過的此神經網路在新的圖像中辨識皮膚。本文最後再將兩種方法的辨識結果比較討論，實驗結果顯示本文所提出的二種方法皆可有效地將大部分皮膚找出，並且也有相當不錯的正確率。 In this article, it makes the analysis discussion of the human skin detection. It detects the main characteristics of skin between non-skin both using RGB colors and chroma components (CbCr) ratios. By selecting a identification criterion, it can distinguish the skin pixels and non-skin pixels in a color image. At the beginning, the detection process will obtain a training image and it manually differentiates into two parts: skin and non-skin pixels. This discrimination process is called as global knowledge. After that different color ratios of these separated skin and non-skin pixels are calculated and their relationships between two parts are analyzed. The first method is using statistical histogram method to detect skin part in a color image. The histograms of different color ratios are derived from the new given image. By selecting proper thresholds, one can makes a best decision to determine whether one pixel is skin or not. The second method is using Backpropagation Neural Network (BPNN) to detect skin. A network is trained by using color ratios obtained above. Then the trained neural network is used to detect a new image and directly classifies the skin and non-skin parts from the given data. Finally the experimental results of two methods are shown and a comparison of two methods has been discussed. Both detection methods are effectively detected the skin pixels in our experiments.
Relation:	華岡工程學報 23期 p.127 -134
Appears in Collections:	[College of Engineering] Chinese Culture University Hwa Kang Journal of Engineering

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	174	View/Open

Loading...