Method for automatically locating an image pattern in digital images using eigenvector analysis

• US 6,625,303 B1
• Filed: 02/09/2000
• Issued: 09/23/2003
• Est. Priority Date: 02/01/1999
• Status: Expired due to Fees

• Current Assignee: CareStream Health Incorporated
• Original Assignee: Eastman Kodak Company
• Inventors: Lee, Hsien-Che, Young, Susan S.
• Primary Examiner(s): Chang, Jon
• Assistant Examiner(s): KIM, CHONG R
• Application Number:
  US09/501,334
• Time in Patent Office: 1,322 Days
• Field of Search: 382/128, 382/107, 382/131, 382/132, 382/190, 382/224, 382/199, 348/169
• US Class Current: 382/132
• CPC Class Codes: A61B 6/505 for diagnosis of bone
• CPC Class Codes: G06T 2207/30008 Bone
• CPC Class Codes: G06T 7/70 Determining position or orientation of objects or cameras camera calibration G06T7/80

Abstract

A method for automatically locating instances of a specified image structure in digital images, comprising the steps of: providing a digital image; detecting simple features associated with the specified image structure in the digital image; for each detected feature, searching, in its spatial neighborhood, for a second or a plural of other features associated with the specified image structure; for each pair of plural of features detected, using an eigenvector classifier to distinguish the wanted features and the unwanted features by matching the eigenvector representation that is constructed from a set of training profiles; and labeling those image regions that are found to be consistent with the specified image structure in the classifying step.

Claims

1. A method of automatically locating cylindrical bones comprising:

providing a digital image having cylindrical bones;

applying a bone edge filter to said digital image;

searching for a pair of candidate starting and ending bone edge points;

determining a profile vector of the line connecting the starting and the ending edge points;

applying an eigenvector classifier; and

saving the classified bone edge points wherein said searching includes;

detecting a starting bone edge point;

generating a line along the gradient direction;

searching for another edge point on the said line in both directions;

determining a profile vector of the line connecting the starting and the ending edge points;

checking the local maxima conditions, that is, for a valid pair of bone edge points, the maximum gradient of the profile is greater than a threshold value g_max; and

checking the local minima conditions, that is, for a valid pair of bone edge points, the minimum gradient of the profile is less than a threshold value g_min.

2. A method of claim 1 wherein said eigenvector classifier is designed by:

providing a set of digital images having cylindrical bones;

determining eigenvector representation of cylindrical bone profiles;

determining a scatter measurement; and

determining bone profile registration.

3. The method of claim 2 wherein said bone profile registration determining includes:

determining horizontal registering;

determining vertical registering; and

determining interpolation.

4. The method of claim 1 including inputting a set of training images;

generating the eigenvector representation of the characteristic profiles of each of said set of training images and using all the data from the generating in applying the eigenvector classifier.