Date of Award

Fall 2007

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical Engineering

Committee Director

Vijayan K. Asari

Committee Member

Zia-ur Rahman

Committee Member

Jiang Li

Call Number for Print

Special Collections LD4331.E55 M33 2007

Abstract

A skin segmentation algorithm robust to illumination changes and skin-like backgrounds is developed in this thesis. So far skin pixel classification has been limited to only individual color spaces and there has not been a comprehensive evaluation of which color components or combination of color components would provide the best classification accuracy, Color components in a given color space form the feature set for the classification of skin pixels. The combination of the color components or the features present within a single color space may not be the best when it comes to skin pixel classification as the discriminatory power of these components to classify the skin pixels is not known. Also, the feature set has to be very small to make the skin segmentation process computationally inexpensive. The Adaboost algorithm has been successfully used for selection of a subset of features required from a given large feature set. Since the data available for skin and non-skin color pixels is large, it is presumed that an ensemble based technique would work better. In this thesis, Adaboost is used for the selection of a best possible subset of color components chosen from different color spaces for the classification of skin pixels and non-skin pixels. This pixel based classification is further improved by taking into account the color information of neighboring pixels. This procedure is motivated by the fact that the property of human skin in which color variations in a given region of human skin are smooth. An image enhancement technique as a preprocessing step is employed to improve the recognition accuracy in images captured in a low or non-uniform lighting environment. The proposed skin segmentation technique is evaluated in images with skin regions of various colors and textures. Its performance is compared with other state-of-the-art techniques and, observed that there is a considerable improvement in recognition accuracy, especially in it is reducing the false positives. Research work is progressing in using the skin segmentation in real life applications such as face detection, face tracking, and ethnicity identification for security purposes.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/x84y-hc24

Download

Share

COinS