Face normalization for recognition and enrollment

• US 8,085,991 B2
• Filed: 03/30/2006
• Issued: 12/27/2011
• Est. Priority Date: 04/14/2005
• Status: Expired due to Fees

• Current Assignee: Rafael - Armament Development Authority Ltd.
• Original Assignee: Rafael - Armament Development Authority Ltd.
• Inventors: Michaeli, Tomer
• Primary Examiner(s): LE, BRIAN Q
• Application Number:
  US11/918,291
• Publication Number:
  US 20090067681A1
• Time in Patent Office: 2,098 Days
• Field of Search: 382/115, 382/118, 382/159, 382/165, 382/170, 382/181
• US Class Current: 382/115
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The present invention is an iterative method for normalization of a probe image against the Eigenspace learned from a database of images. The invention is also an iterative method for normalizing the n images in a database, wherein the normalization is carried out without using a predetermined criterion.

Claims

1. An iterative method for normalization of a probe image against the Eigenspace learned from a database of images, said method comprising the following steps:

(a) initialize the transformation parameters according to said probe image gathered in the detection stage;

(b) construct a deformed grid using said transformation parameters;

(c) resample said image using said deformed grid to obtain a resampled image;

(d) form a set of linear equations that describe the set of transformation parameters and Eigenspace coefficients that, if used to resample the image, will allow improved representation of the image in the Eigenspace as compared to the representation obtained in the previous iteration;

(e) solve said set of linear equations to simultaneously obtain said transformation parameters and said Eigenspace coefficients; and

(f) repeat steps (b) through (e) until the change in said transformation parameters between two consecutive iterations are less than a predetermined threshold.

2. An iterative method for normalizing the n images in a database without using a predetermined criterion, said method comprising the following steps:

(a) construct an Eigenspace comprising all of the images in said database except one;

(b) use the method of claim 1 to normalize the one image with respect to the Eigenspace formed from the remaining n−

1 images;

(c) repeat steps (a) and (b) n times, each time for a different image; and

(d) repeat steps (a) to (c) using said normalized images until the changes in the transformation parameters between two consecutive iterations are smaller than a predetermined value.