Apparatus for extracting feature points from a facial image

• US 5,905,807 A
• Filed: 09/05/1996
• Issued: 05/18/1999
• Est. Priority Date: 01/23/1992
• Status: Expired due to Term

• Current Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Kado, Yoshiyasu, Nakagawa, Masamichi, Maehara, Fumio
• Primary Examiner(s): Au, Amelia
• Assistant Examiner(s): Prikockis, Larry J.
• Application Number:
  US08/708,873
• Time in Patent Office: 985 Days
• Field of Search: 382/118, 382/115, 382/190, 382/195, 382/197, 382/199, 382/261, 382/270
• US Class Current: 382/118
• CPC Class Codes: A61B 5/1176 Recognition of faces
• CPC Class Codes: G06V 40/168 Feature extraction; Face representation
• CPC Class Codes: H04N 19/20 using video object coding

Abstract

A facial image picture data is processed by an edge extraction part, which performs edge extraction on the inputted image picture. The processed data is then converted by a binary level conversion part for each preestimated point of facial elements. An image picture arithmetic processing part performs time-to-time arithmetic computation of correlation values between the data of the binary-leveled edged-image picture and the data having been previously stored in the shape data-base part, and issues an output including extracted data of feature points of the facial image picture.

Claims

1. An apparatus for outputting data indicative of matching features of a face with data in a data-base, said apparatus comprising:

an image picture inputter for receiving an input image of a face, said image including a plurality of facial feature searching regions each of which corresponds to a respective facial feature of said face;

an edge extractor for processing said input image of said face to form an edged-image of said face;

a binary level converter for dividing said image into said plurality of facial feature searching regions and subsequently performing a binary level conversion on said edged-image of said face corresponding to a plurality of facial feature points of said face, said binary level converter converting said edged-image of said face into a binary-leveled edged-image of said face;

a shape data-base permanently storing a plurality of shape data as a reference data-base for said plurality of facial feature points, said shape data comprising coordinate data and gradient vectors in x,y coordinates;

an image picture arithmetic processor for determining an amount of correspondence by computing a correlation value between data of each of said plurality of facial feature points of said binary-leveled edged-image of said face and each corresponding one of said plurality of facial feature points of said shape data stored in said shape data-base;

a shape data-base changer for changing copied data of said plurality of shape data copied from said shape data-base; and

an outputter for outputting data indicative of said matching features of said face based on said correlation value and said copied data, wherein

using both said edge extractor and said binary level converter results in decreased sensitivity of said apparatus to changes in lighting.

2. An apparatus in accordance with claim 1, said apparatus further comprising:

a facial feature region determiner for determining locations of others of said plurality of facial feature searching regions based on said data output from said outputter with respect to a determination of a first facial feature.

3. An apparatus in accordance with claim 2, wherein:

a region of irises of said face is determined before said facial feature region determiner determines said locations of said others of said plurality of facial feature searching regions.

4. An apparatus in accordance with claim 2,

further comprising a second binary level converter arranged to perform binary level conversion before said processing of said input image of said face by said edge extractor.

5. An apparatus in accordance with claim 4, when said:

second binary level converter is connected to said edge extractor and is arranged to receive an output of said facial feature region determiner, to divide one of said plurality of facial feature searching regions into a plurality of smaller sub-regions, and to determine a binary level conversion threshold value for each of said plurality of smaller sub-regions from a brightness distribution of each of said plurality of smaller sub-regions.

6. Amended) An apparatus in accordance with claim 1, said apparatus further comprising:

an adoption/rejection selector for obtaining expected positions of said plurality of facial features, which obtains and determines an accuracy of positions of said plurality of facial features, and which discards any erroneously extracted facial features prior to processing said input image of said face by said edge extractor.

7. An apparatus in accordance with claim 1, wherein:

said edge extractor includes a computer using a Sobel operator which produces gradient vectors for respective pixels of said image of said face, said gradient vectors having magnitudes and directions indicative of said facial features.

8. An apparatus in accordance with claim 1, wherein:

an arithmetic computation performed by said image picture arithmetic processor is an inner product between compared gradient vectors of said binary-leveled edged-image of said face and respective shape data.

9. An apparatus in accordance with claim 8, wherein:

a location of at least one of said plurality of facial feature searching regions is determined by a location on said image of said face where said correlation value is largest.

10. An apparatus in accordance with claim 3, wherein:

said respective locations of said others of said plurality of facial feature searching regions are determined with reference to a position of said region of said irises of said face.

11. An apparatus in accordance with claim 1 wherein said shape data-base changer changes said copied data by a method comprising:

(1) displacing said coordinate data by ±

1 in a direction of said gradient vector,

(2) changing said direction of said gradient vector after said step of displacing said coordinate data, so as to match said shape data at said displaced coordinate data,

(3) repeating steps (1) and (2) by displacing said coordinate data further by ±

1 thereby pursuing an increase in value of said correlation value, and

(4) ceasing said repeating step when one of said value of said correlation value no longer increases and said value of said correlation value exceeds a predetermined threshold value.

12. An apparatus in accordance with claim 1, wherein said shape data comprises:

said coordinate data which are defined as;

P_k =(l_k, m_k), where l_k, m_k, are x, y coordinates of a k-th position at which said shape data is given, respectively, and

said gradient vectors are defined as;

v_k =(v_x, v_y), where

v_x² +v_y² =1 and

1≦

k≦

n (where n is a number of positions).

13. An apparatus in accordance with claim 2, wherein said shape data comprises:

said coordinate data which are defined as;

P_k =(l_k, m_k), where l_k, m_k, are x, y coordinates of a k-th position at which said shape data is given, respectively, and

said gradient vectors are defined as;

v_k =(v_x, v_y), where

v_x² +v_y² =1 and

1≦

k≦

n (where n is a number of positions).