Method of correcting deviation of detection position for human face, correction system, and correction program

• US 7,415,140 B2
• Filed: 01/26/2005
• Issued: 08/19/2008
• Est. Priority Date: 01/27/2004
• Status: Expired due to Fees

• Current Assignee: Seiko Epson Corporation (Seiko Group)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Nagahashi, Toshinori, Hyuga, Takashi
• Primary Examiner(s): Ahmed; Samir A.
• Assistant Examiner(s): Woldemariam; Aklilu k
• Application Number:
  US11/043,900
• Publication Number:
  US 20050180612A1
• Time in Patent Office: 1,301 Days
• Field of Search: 382/118
• US Class Current: 382/118
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

A method of correcting the deviation of a position at which a human face is detected using a face detection frame F. An image G within the detection frame F is divided left and right along the portion of the center axis C. The degree of similarity of mirror-symmetrical relationship between the image portions G-1 and G-2 obtained by the division is calculated. An amount of positional deviation d that maximizes the degree of similarity and its direction are found. The frame F is moved a distance corresponding to the amount of positional deviation d in the direction. Thus, the deviation of the detection position due to the detection frame F can be corrected by a quick and simple method.

Claims

1. A method of correcting deviation of a detection position for a human face detected using a face detection frame, said method comprising the steps of:

dividing an image within said face detection frame into left and right image portions about a central axis portion thereof, the image within said face detection frame having an odd number of pixels in a left-and-right direction;

adjusting the number of pixels in the left-and-right direction of the image within said face detection frame to an even number by moving at least one of left and right ends of the face detection frame in a left or right direction so as to stretch or contract the face detection frame;

calculating a degree of similarity of mirror-symmetrical relationship between the two image portions based on a correction coefficient obtained based on a value of a pixel from the left image portion and a value of a corresponding pixel from the right image portion;

finding an amount of positional deviation at which the calculated degree of similarity maximizes and a direction of the positional deviation; and

thereafter moving said face detection frame in said direction according to the amount of positional deviation.

2. A method of correcting deviation of a detection position for a human face as set forth in claim 1, wherein the degree of similarity of mirror-symmetrical relationship between said image portions obtained by the division is determined based on the correlation coefficient utilizing edge intensities of all pixels constituting the image within said face detection frame.

3. A method of correcting deviation of a detection position for a human face as set forth in claim 1, wherein said face detection frame is a rectangular form parallel to a center line of said human face.

4. A method of correcting deviation of a detection position for a human face as set forth in claim 2, wherein average values of edge intensities of the pixels taken in an up-and-down direction are used as the edge intensities.

5. A method of correcting deviation of a detection position for a human face as set forth in claim 1, wherein said face detection frame is set using a face detection technique.

6. A method of correcting deviation of a detection position for a human face as set forth in claim 2, wherein said edge intensities use a Sobel edge detection operator.

7. A system for correcting deviation of a detection position for a human face detected using a face detection frame, said system comprising:

an image reading portion for reading an image containing the human face;

a face detection frame setting portion for setting the face detection frame in a portion of the human face within the image read by said image reading portion;

an image division portion for dividing the image within the face detection frame detected by said face detection frame setting portion into left and right image portions about a central axis thereof, the image within the face detection frame having an odd number of pixels in a left-and-right direction, the image division portion adjusting the number of pixels in the left-and-right direction of the image within the face detection frame to an even number by moving at least one of left and right ends of the face detection frame in a left or right direction so as to stretch or contract the face detection frame;

a degree of similarity calculation portion for calculating a degree of similarity between the left and right image portions obtained by said image division portion based on a correction coefficient obtained based on a value of a pixel from the left image portion and a value of a corresponding pixel from the right image portion;

a positional deviation amount calculation portion for calculating an amount of positional deviation at which the degree of similarity obtained by said degree of similarity calculation portion maximizes and a direction of the positional deviation; and

a position correction portion for moving said face detection frame a distance in said direction, the distance corresponding to the amount of positional deviation calculated by said positional deviation amount calculation portion.

8. A system for correcting deviation of a detection position as set forth in claim 7, wherein said degree of similarity calculation portion calculates the correlation coefficient about mirror-symmetrical relationship between said left and right image portions based on edge intensities of all pixels within each of the image portions.

9. A computer readable medium storing a program for correcting deviation of a detection position for a human face detected using a face detection frame, said program acting to cause a computer to perform functions of the system of claim 7.

10. The computer readable medium storing a program for correcting deviation of a detection position as set forth in claim 9, wherein said degree of similarity calculation portion calculates the degree of similarity according to the correlation coefficient about mirror-symmetrical relationship between said left and right image portions by making use of edge intensities of all pixels within each of said image portions.