Skip to Main content Skip to Navigation
Theses

Segmentation région-contour des contours des lèvres

Abstract : This report presents the thesis that has been jointly conducted at the Grenoble Institute of Technology in France and at the Laval University in Canada. The work involves the GIPSA-lab in Grenoble and the CVSL of the Laval University. The aim of the thesis is to propose a set of robust methods to segment the lips outer and inner contours. In terms of robustness, we intend to propose a reliable lips contours segmentation that does not require the tuning of parameters. In the fist chapter a state of the art of mouth analysis is given. In the second chapter of this report, we will introduce our “region-contour” based approach to segment a binary mask of the lips on static face color images. First, we will describe the decorrelation-stretch algorithm we use to enhance the contrast between lips pixels and skin pixels and then a multi-scale gradients approach to contour modeling. The last part of the chapter will focus on the segmentation of a lips binary mask by automatic thresholding of a specific chromatic component. In chapter three we will be interested in the mouth state identification problem. The knowledge of mouth state is critical if one is to propose a robust segmentation of the lips internal contours. A bio-inspired approach based on retina and visual cortex models has been developed to compute a scale invariant mouth description: the log-polar spectrum. Log-polar spectrums, computed on a manually classified mouth images database, are used to train a SVM network. The goal of the network is to classify unknown mouth images in 2 clusters: open mouth and closed mouth. Recently, infrared based approaches have become popular in face analysis, especially for face recognition problems. Infrared thermography is an area of extensive expertise at the CVSL. In order to enhance the robustness of the lips contours segmentation, we studied the potential of the information given by infrared face images. A combined visible/infrared face image database has been constructed for that purpose. Chapter 4 describes the construction of the combined visible/infrared database and the statistical study of the skin/lips contrast on infrared face images. The last chapter deals with the outer and inner lips contours segmentation. The mouth outer contour is processed first. Using the lips binary mask given by the algorithm described in chapter 2, we have developed an adaptive contour model for the external mouth contour. The contour will be described by a polynomial curve. The complexity of the curve will be adapted using color and multi-scale gradients information. For the inner lips contour, we proposed an unsupervised classification method to segment the inner areas of the mouth. This gives us a binary mask of the inner areas of the mouth. Finally, given the mask of the inner areas of the mouth, the contour is extracted by using the same method that has been applied on the outer contour. The last section of chapter 5 will deal with the performance evaluation of our segmentation algorithms. An original approach for contour comparison, based on Fourier descriptors, has been developed for that purpose.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00462022
Contributor : Christian Bouvier <>
Submitted on : Monday, March 8, 2010 - 11:22:07 AM
Last modification on : Thursday, November 19, 2020 - 12:59:40 PM
Long-term archiving on: : Wednesday, March 29, 2017 - 11:36:20 PM

Identifiers

  • HAL Id : tel-00462022, version 1

Collections

Citation

Christian Bouvier. Segmentation région-contour des contours des lèvres. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 2010. Français. ⟨tel-00462022⟩

Share

Metrics

Record views

817

Files downloads

3962