×

Automatic skin color model face detection and mean-shift face tracking

  • US 7,869,631 B2
  • Filed: 12/11/2006
  • Issued: 01/11/2011
  • Est. Priority Date: 12/11/2006
  • Status: Active Grant
First Claim
Patent Images

1. A method of automatically detecting and tracking a face by an electronic capturing device, comprising:

  • executing a face detecting algorithm to detect a face position in a frame in a red, green and blue (RGB) chromaticity space based on a skin color model, wherein said face detecting algorithm performs a color transformation for transforming said frame into a new frame in a hue, saturation and value (HSV) chromaticity space and comprises steps of;

    calculating a histogram of hue of color pixels on said new frame when r>

    max(g,b, S) and S>

    38, wherein S represents saturation of said new frame, and r>

    max(g,b,S) and S>

    38 represents that intensity of red color of said frame is larger than maximum value of intensities of green color and blue color of said frame when the saturation S of said new frame is larger than 38;

    obtaining a maximum value Maxh of said histogram of hue when number of said color pixels multiplied by 256 is larger than total number of all pixels on said new frame;

    obtaining a smallest value i of said histogram of hue, such that said pixels of said new frame with hues falling within a histogram hue range [Maxh−

    i, Maxh+i] are determined to be skin color pixels respectively;

    forming a mask for said skin color pixels;

    marking a label for all continuous regions in said new frame for defining a rectangle corresponding to said label when a number of pixels of said continuous regions is larger than the total number of pixels on said new frame divided by 256;

    using a face feature matching engine to analyze and determine whether an image within said rectangle is a face or not;

    generating a detected rectangle for covering at least one rectangle;

    calculating a skin color probability density function for an image within said detected rectangle; and

    outputting information of detecting a face within said detected rectangle and displaying said detected rectangle on said new frame corresponding to said face; and

    executing a face tracking algorithm to detect face positions in subsequent new frames by using a nonparametric technique and a mean shift algorithm according to said face.

View all claims
  • 5 Assignments
Timeline View
Assignment View
    ×
    ×