Image analysis – Image transformation or preprocessing – Changing the image coordinates
Reexamination Certificate
2000-02-22
2004-01-20
Do, Anh Hong (Department: 2624)
Image analysis
Image transformation or preprocessing
Changing the image coordinates
C382S190000, C382S302000
Reexamination Certificate
active
06681057
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to the field of image registration, and more particularly to an image registration system implementing PID control techniques.
DESCRIPTION OF THE RELATED ART
In general, registration may be defined as the process by which the correspondence between two or more arrays of values is performed. Image registration may be defined as the process by which the correspondence between a subset or all points in two or more images, typically of the same scene, is determined. Image registration may also be stated as the process of estimating the affine transformation between the images. The affine transformation may be defined as the rotation, scale change, and translation that maps one n-dimensional image into a second n-dimensional image. In practice, n is usually 2 or 3. Image registration usually plays an important and critical first step in applications involving the fusion of information from multiple modalities. Modalities refer to various sensor implementations to receive the images or array of values. For example, an input image may be received by a thermal camera; whereas, a reference image may be received by a regular camera. An input image may exhibit an affine transformation with respect to the reference image.
Image registration may be used in image analysis tasks such as motion or change detection, fusion of data from multiple sensing modalities, and image geometric correction. There has been a tremendous increase in the need for good image registration techniques due to the increased use of temporal and multimodal 2-D and 3-D images in medical, remote sensing, and industrial applications.
Prior image registration techniques may have implemented some notions of feedback control systems. However, prior image registration techniques did not use PID (P-proportional, I-integration, D-differential components) control techniques to determine the affine transformation of one image with respect to another image.
Background of classical control systems is deemed appropriate. Control systems at the most basic level deal with a system that has a specific input which generates a specific output. The system could be a plant, an engine, a biological object, or any other structure with a clear input—output behavior. Inputs to the control system may be interpreted as control actions or noise sources. Outputs of a given control system mainly comprise measurements. These measurements describe internal states, which, in most cases, are not directly measurable.
In the simplest case, a so-called set point must be reached and stabilized based on appropriate control actions. Even in highly nonlinear situations, linear models can successfully be used to determine such control strategies. For example, the general continuous linear model
ⅆ
x
(
t
)
ⅆ
t
=
Ax
(
t
)
+
Bu
(
t
)
y
(
t
)
=
Cx
(
t
)
+
Du
(
t
)
(1)
can be described by:
matrices A, B, C, and D
vector x(t) of internal states
vector u(t) of control actions
vector y(t) of measurements.
Additionally, noise terms can be added if necessary. The most successful and very common control strategy is simply a feedback according to:
u
(
t
)=−
K
(
t
)
x
(
t
) (2)
The design of a valid control system mainly comprises the construction of matrices K(t) making the control system as efficient as possible. In typical applications the matrix K(t) is constant which simplifies the algorithms considerably.
Prior image registration techniques implementing notions of feedback strategies were deficient in that the convergence rate to determine the affine transformation was unsatisfactory. Therefore, an improved system and method is desired for performing registration, e.g. determining the affine transformation of an input image with respect to a reference image.
SUMMARY OF THE INVENTION
A system and method are provided for improving the accuracy and convergence rate in determining the affine transformation of one image or array of values with respect to another image or array of values. According to one embodiment of the present invention, the problem of finding the affine transformation parameters between the images or array of values to be registered is posed as a classical control problem using PID (P-proportional, I-integration, D-differential components) techniques. According to one embodiment, registration may be performed between an input array of values and a reference array of values. According to another embodiment, image registration may be performed between an input image and a reference image. An efficient registration or image registration may be performed by estimating the transformation between two subsets using a strategy based on a PID control system.
The present invention comprises a system, method and memory medium for improved registration, such as image registration. In a first embodiment, the system comprises a PID (P-proportional, I-integration, D-differential components) controller to accurately determine the affine transformation. Hence the PID controller may be used to determine the amount of rotation, shift and scale of an input image relative to a reference image. Both the reference and input image may be comprised of pixels and may be functions of position, angle and scale. In another embodiment, the PID controller may be used to determine the amount of rotation, shift and scale of an input array of values relative to a reference array of values. The reference and input array of values may both be functions of position, angle and scale.
In a particular embodiment, the reference image may be preprocessed by constructing a representation of the reference image, preferably a gradient matrix, &lgr;, comprising gradient information of the position values of the reference image. The input image may be input to an image registration system.
An estimate may then be constructed comprising initial estimates of at least one of a position, angle and scale value of the input image. Preferably, the estimate is an estimate matrix, &rgr;.
The input image may then be compared with the reference image to determine if the input image has been shifted, rotated or scaled with respect to the reference image. More specifically, an error may then be calculated by subtracting the input image from the reference image. The error is preferably represented by an error matrix, e.
A comparison may subsequently be made between the error matrix and a given value. If the error matrix is less than a given value, then the correspondence between the input image and the reference image has been determined. Otherwise, the process continues by multiplying the error matrix, e, with the gradient matrix, &lgr;, and the estimate matrix, &rgr;. The result of the multiplication is a new change in the position, angle and scale values of the input image. The input image is consequently transformed by replacing the old values of the position, angle and scale with the new values of the position, angle and scale. These new values are constructed from the changes in the position, angle and scale values of the input image. For example,
X
new=
X+K*&Dgr;X+Ki*∫&Dgr;X+Kd*d/dx
(&Dgr;
X
)
Y
new=
Y+K*&Dgr;Y+Ki*∫&Dgr;Y+Kd*d/dy
(&Dgr;
Y
)
angle new=angle+
K
*&Dgr;angle+
Ki
*∫&Dgr;angle+
Kd*d/d
angle(&Dgr;angle)
scale new=scale+
K
*&Dgr;scale+
Ki
*∫&Dgr;scale+
Kd*d/d
scale(&Dgr;scale).
The integral constant, Ki, is a steady state error correction parameter that is multiplied to the average of the previous values of the change in position, angle and scale. The derivative constant, Kd, is preferably a response parameter which may reduce the swings in the estimate values from being too high or too low. K is a constant that is multiplied to &Dgr;X, &Dgr;Y, &Dgr; angle, and &Dgr; scale. &Dgr;X is the change in the value of X of the input image. &Dgr;Y is the change in the value of Y of the input image. &Dgr; angle is the change in the angle value of the input image. &D
Nair Dinesh
Wenzel Lothar
Do Anh Hong
Hood Jeffrey C.
Meyertons Hood Kivlin Kowert & Goetzel P.C.
National Instruments Corporation
LandOfFree
Image registration system and method implementing PID... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image registration system and method implementing PID..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image registration system and method implementing PID... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3194036