Browse Prior Art Database

Efficient Skin Color Extraction System

IP.com Disclosure Number: IPCOM000168514D
Original Publication Date: 2008-Mar-26
Included in the Prior Art Database: 2008-Mar-26
Document File: 5 page(s) / 301K

Publishing Venue

Siemens

Related People

Juergen Carstens: CONTACT

Abstract

In the fields of gesture recognition, face detection, image content filtering and human motion capturing a system for extracting skin colors within changing illumination conditions in real time and for implementations with low processing power and small storage capacity is not known up to the present moment. Two commonly used models are Parametic and Non-Parametric Skin Distribution Modeling, both demanding high performance and a great amount of memory capacity. These disadvantages limit the adequacy for implementations with high requirements and limitations on computer resources like implementations on a DSP (Digital Signal Processor). For better understanding of the proposed solution, the two commonly used methods are described shortly in the following: a) Non-Parametric Skin Distribution Modeling

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 51% of the total text.

Page 1 of 5

Efficient Skin Color Extraction System

Idea: Suresh Kumili, IN-Bangalore

In the fields of gesture recognition, face detection, image content filtering and human motion capturing a system for extracting skin colors within changing illumination conditions in real time and for implementations with low processing power and small storage capacity is not known up to the present moment.

Two commonly used models are Parametic and Non-Parametric Skin Distribution Modeling, both demanding high performance and a great amount of memory capacity. These disadvantages limit the adequacy for implementations with high requirements and limitations on computer resources like implementations on a DSP (Digital Signal Processor). For better understanding of the proposed solution, the two commonly used methods are described shortly in the following:
a) Non-Parametric Skin Distribution Modeling
The key idea of this approach is to estimate the skin color distribution from the training data without deriving an explicit model of the skin color. The result of the method is sometimes referred to as construction of Skin Probability Map (SPM), where a probability value to each point of a discrete color value is assigned.

The disadvantage is the great amount of memory required to store the skin probabilities. For example, considering RGB (Red Green Blue) quantized to 8 bits per color, an array of 224 (=16,777,216) elements is required to store the skin probabilities.
b) Parametric Skin Distribution Modeling
The central method of this approach is to reflect skin color in different color-areas like e.g. HSV (Hue, Saturation, Value) in which the skin model is quite compact. Different models are known like the Elliptic Boundary Model, Single Gaussian and Mixture of Gaussians.

A disadvantage is the high demand of computational performance needed for solving the equations of the approach. In the following, the equations for the Gaussian Model in HSV color-area are shown:

Obviously, the three equations (1), (2) and (3) for the HSV color-area are quite demanding for performance capacities with summarized seven additions/subtractions, seven multiplications/divisions and three complex operations (square-root, min, arcos). Equation (4) represents the Single Gaussian Model and must be evaluated for every pixel, which is obviously also quite computationally intensive. A new procedure for extracting skin colors within changing illumination conditions in real time and for implementations with low processing power and small storage capacity is proposed in the following. The idea consists of two steps: Training and Testing.

Training: A database of skin colors is obtained. In this example the data is received from the Internet. All sample skin colors are converted from RGB to rgb (normalized RGB) with the following equations:

© SIEMENS AG 2008 file: 2007J26524.r...