Dismiss
IP.com applications will be updated on Sunday, March 5, from 11 am to 2 pm ET, to add new functionality and content. You may experience brief service interruptions during this period. We apologize for any inconvenience.
Browse Prior Art Database

Method for face detection based on adaptive color segmentation

IP.com Disclosure Number: IPCOM000007119D
Publication Date: 2002-Feb-26
Document File: 9 page(s) / 466K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for face detection based on adaptive color segmentation. Benefits include improved color detection and processing and improved performance.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 19% of the total text.

Method for face detection based on adaptive color segmentation

Disclosed is a method for face detection based on adaptive color segmentation. Benefits include improved color detection and processing and improved performance. 

Background

              A central task in many computer vision applications, such as face encoding, face recognition, face tracking, and face animation, is to detect the facial region of input images with a complex background.

              Human skin is known to have a specific color distribution. Human skin color has been proven to be an effective feature in many applications for human face detection and tracking. Color is the simplest attribute of the set of pixels that make up an image. Typically, color is quantitatively represented as a three-dimensional vector.

              Several variations of the statistical method for face segmentation use color information in different color spaces. These methods can be described in three basic groups. The first group of the methods is based on using linear discriminates. The ranges of color values in some color planes for the skin color using color histograms from training images are predefined. Then these values are explored to mark pixels with color values within these ranges in input images.

              Another group of methods uses color histogram statistics more completely. This group is found in color distribution models. A typical face color histogram is created. Then, the histogram of each image area to be classified is projected on this histogram. A great deal of work is based on Gaussian models. As a rule, all of the above methods of color segmentation require preliminary training and definition of the great number of parameters that characterize skin-color distribution.

              The third group of methods of color segmentation is based on neural networks. However, these systems are computationally very demanding. Therefore, they are not very applicable to applications where real-time or near real-time performance is required.

General description

              The disclosed method detects faces on an input video sequence from a color video camera. This method is based on an adaptive color segmentation algorithm. It includes the following stages performed on every frame of the video sequence:


1.           Color segmentation using 2D color histograms in HSV or YCrCb color spaces
2.           Iterative refinement of the face mask
3.           Reference histogram adaptation

              For face segmentation on the first frame of the video sequence, a typical skin-color histogram is used. This histogram is calculated from training images. On all subsequent frames, the reference histogram is updated by segmentation results on the previous frame.

              The disclosed method is characterized by the high speed of the processing and by the high segmentation stability because of the use of iterative refinement and histogram adaptation.

Advantages

              The disclosed method presents several advantages over conventional methods. The color segmentation algorithm of the disclosed method does not require exten...