Method for the automatic resetting of images

• US 5,048,103 A
• Filed: 03/16/1990
• Issued: 09/10/1991
• Est. Priority Date: 03/29/1989
• Status: Expired due to Fees

• Current Assignee: General Electric CGR SA (General Electric Company)
• Original Assignee: General Electric CGR SA (General Electric Company)
• Inventors: Lienard, Jean, Leclerc, Vincent
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Fallon, Steven P.
• Application Number:
  US07/494,395
• Time in Patent Office: 543 Days
• Field of Search: 382/6, 382/44, 358/111, 358/20, 364/413.13, 364/414
• US Class Current: 382/294
• CPC Class Codes: G06T 3/153 using elastic snapping
• CPC Class Codes: G06T 3/40 Scaling of whole images or parts thereof, e.g. expanding or contracting
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 5/80 Geometric correction
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches

Abstract

Images representing one and the same scene, but acquired under conditions such taht the subject has moved, are automatically reset by making an automatic selection of a homogenous population of landmarks and by computing the shift vectors associated with these landmarks. It is shown that, in order to select the landmarks, it is worthwhile to characterize the image by a certain number of characteristic mangitudes representing the solidity of these landmarks. The solidity of the landmarks is the ability of a landmark to be the site of an exact measurement of shift for a required precision.

Claims

1. A method for the automatic resetting of images wherein a resetting is done of a second image, called a mask image, distorted with respect to a first image, called a reference image, each image having points and being defined as a collection of addresses of these points, each address being associated with a gray level, said method including the following steps:

a selection is made, preferably in the first image, of a set of landmarks;

a search is made for the correspondents of these landmarks in the second image;

an estimation is made of the shifts undergone by the selected landmarks of the first image with respect to the corresponding searched-for landmarks of the second image;

one of the images, preferably the second image, is restored by correcting the position or the gray level of the points of this image in taking into account the shifts of the landmarks located in the neighborhood of these points;

wherein, to make the selection,

an assigning is done, to points of the first image, of a computed magnitude or of computed magnitudes characteristic of the solidity and of the homogenity of distribution of points of this image liable to act as landmarks;

a sorting-out is done of the landmarks to be chosen as a function of the meeting of a criterion by a combination of the magnitudes at these points;

wherein, to do the sorting out,

a computation is made, at these points, of at least one of the following five magnitudes;

*X₁ ;

the difference, preferably standardized, between the two images in the means of the gray levels, these means being computed in each of the two images on a population of points contained in windows framing considered superimposable points;

*X₂ ;

the standard deviation of the gray levels of a population of points located in a window framing, in the first image, a landmark to be characterized;

*X₃ ;

the useful value of Moravec for the points, to be characterized, of the first image;

*X₄ ;

the standardized value of the gray level gradient, computer on a landmark to be characterized in the first image;

*X₅ ;

the value (a) of the damping coefficient of the self-correlation function modelized by an exponential function, at the position of the landmark to be characterized.

2. A method according to claim 1 wherein the landmarks are sorted out as a function of a combination of magnitudes, this combination being expressed analytically as a function of these magnitudes, coefficients B_O, B₁, B₂, B₃, B₄ and B₅ of this combination being determined by a statistical, analytical assessment on a population of images of one and the same type, the characteristics of restoration of which are already known.

3. A method according to either of claims 1 or claim 2, wherein, wherein the coefficients have the following values:

B_O =0.5, B₁ =4.8 and B₅ =-2.7 when the combination concerns the two magnitudes X₁ and X₅.

4. A method according to claim 1 wherein:

the landmarks are selected hierarchically, according to a quadtree method, in performing an iterative sub-sampling (M+1) of the landmarks of the first image selected for a given resolution (M+2) of this first image.

5. A method according to claim 4, wherein:

an assessment is made, at the sub-sampled level, of the distribution of the landmarks and of their solidity, and

the landmarks are selected by choosing a minimum of an analytical function of transition cost for which this distribution and this solidity are the variables.

6. A method according to either of the claims 1 or 2 wherein, to make the selection, one and the same number of landmarks is chosen for each window of a set of related windows covering the first image.

7. A method according to claim 6, wherein a second sub-sampling, following the first one, is done by filtering the consistent landmarks for which the associated shift vector, computed by another way, is the most coherent with respect to its neighbors.

8. A method according to claim 1 wherein the shifts are estimated by cross-correlation and wherein the fractional shifts involved are sought by approximation by a quadric.