Title: Active Shape Models on adaptively refined mouth emphasizing color images
Authors: Panning, Axel
Al-Hamadi, Ayoub
Michaelis, Bernd
Citation: WSCG 2010: Communication Papers Proceedings: 18th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision in co-operation with EUROGRAPHICS, p. 221-228.
Issue Date: 2010
Publisher: Václav Skala - UNION Agency
Document type: konferenční příspěvek
URI: http://wscg.zcu.cz/WSCG2010/Papers_2010/!_2010_Short-proceedings.pdf
ISBN: 978-80-86943-87-9
Keywords: extrakce znaků;segmentace úst;zpracování obrazů;aktivní modely tvarů
Keywords in different language: feature extraction;lip segmentation;image processing;active shape models
Abstract: In this paper, we propose a hybrid method for lip segmentation based on normalized green-color histogram splitting and Active Shape Models (ASM). A new adaptive method for histogram splitting is applied in two steps. First, after defining a region of interest for mouth segmentation, a rough adaptive threshold selects a histogram region assuring that all pixels in that region are skin pixels. Second, based on these pixels, we build a Gaussian model which represents the skin pixels distribution and is used to obtain a refined optimal threshold for lip pixel classification. This process is used to refine the normalized green channel image for the elimination of inner distortions and gradients inside the lip region, which can misguide active contours (i.e. ASM) in the last step of the hybrid segmentation process. In the results, we present that the proposed method performed better than conventional ASM on unrefined color enhanced images or pure color-histogram based mouth segmentation.
Rights: © Václav Skala - UNION Agency
Appears in Collections:WSCG 2010: Communication Papers Proceedings

Files in This Item:
File Description SizeFormat 
Panning.pdfPlný text866,06 kBAdobe PDFView/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/11063

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